diff --git "a/attnserver.run_attnserver.slurm.sh.343209.out.log" "b/attnserver.run_attnserver.slurm.sh.343209.out.log" --- "a/attnserver.run_attnserver.slurm.sh.343209.out.log" +++ "b/attnserver.run_attnserver.slurm.sh.343209.out.log" @@ -535,3 +535,13338 @@ make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/mega > compiling and loading fused kernels ... INFO:megatron.training.initialize:Setting logging level to 0 >>> done with compiling and loading fused kernels. Compilation time: 2.457 seconds +time to initialize megatron (seconds): 7.043 +[after megatron is initialized] datetime: 2025-06-21 21:19:52 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 74511872 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 74511872 + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 74511872 + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 74511872 + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 74511872 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 74511872 + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 74511872 + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 74511872 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (74511872 elements, 74511872 padded size): + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.decoder.final_layernorm.bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.final_layernorm.weight + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.embedding.position_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc1.bias +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (2.06, 3.81) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:19:52 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=1024, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.007080 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 66592 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.003368 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 66562 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.003269 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 66686 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:19:52 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (261.26, 284.68) + train/valid/test-data-iterators-setup ..........: (35.72, 138.47) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:19:52 +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor:batch tensor: tokens tokens torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096])batch tensor: + batch tensor:labels loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: batch tensor: loss_maskattention_mask torch.Size([4, 4096]) +torch.Size([4, 1, 4096, 4096]) +batch tensor:batch tensor: attention_maskposition_ids torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp:batch tensor after cp: labelstokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: loss_mask labels torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:attention_mask loss_mask torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: position_idsattention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:20:02] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 9280.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 5] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3320.0 | max reserved: 3320.0[Rank 4] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3448.0 | max reserved: 3448.0 + +[Rank 3] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3320.0 | max reserved: 3320.0 +[Rank 6] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3320.0 | max reserved: 3320.0 +[Rank 7] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3448.0 | max reserved: 3448.0 +[Rank 1] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3704.0 | max reserved: 3704.0 +[Rank 2] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3512.0 | max reserved: 3512.0 +[Rank 0] (after 1 iterations) memory (MB) | allocated: 1055.16357421875 | max allocated: 3235.26806640625 | reserved: 3320.0 | max reserved: 3320.0 +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_maskbatch tensor: torch.Size([4, 1, 4096, 4096]) + batch tensor:tokens position_ids torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor:batch tensor after cp: loss_mask torch.Size([4, 4096]) +tokensbatch tensor after cp: attention_mask batch tensor after cp:torch.Size([4, 1, 4096, 4096])batch tensor: +tokenstorch.Size([4, 4096])batch tensor: +batch tensor after cp:tokens batch tensor: position_ids torch.Size([4, 4096]) +tokenslabelsbatch tensor after cp:torch.Size([4, 4096])torch.Size([4, 4096]) +labels +torch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor:batch tensor: +torch.Size([4, 4096]) +batch tensor after cp:loss_masklabels loss_maskbatch tensor:torch.Size([4, 4096]) torch.Size([4, 4096])labels + +torch.Size([4, 4096]) batch tensor: +torch.Size([4, 4096])batch tensor: attention_maskbatch tensor after cp: +loss_mask batch tensor:torch.Size([4, 1, 4096, 4096])attention_mask + torch.Size([4, 4096])loss_masktorch.Size([4, 1, 4096, 4096]) +batch tensor: + batch tensor:batch tensor after cp:position_idstorch.Size([4, 4096]) +attention_maskposition_idstorch.Size([4, 4096]) batch tensor: +torch.Size([4, 1, 4096, 4096]) torch.Size([4, 4096]) +attention_mask + batch tensor:torch.Size([4, 1, 4096, 4096]) +position_ids batch tensor:torch.Size([4, 4096]) +position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:labels tokenstorch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: +batch tensor after cp: batch tensor after cp:loss_mask tokens labels torch.Size([4, 4096])torch.Size([4, 4096])torch.Size([4, 4096]) + + +batch tensor after cp:batch tensor after cp: batch tensor after cp: labelsattention_maskloss_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: +batch tensor after cp: batch tensor after cp: loss_mask position_ids attention_masktorch.Size([4, 4096])torch.Size([4, 4096]) + +torch.Size([4, 1, 4096, 4096])batch tensor after cp: + batch tensor after cp:attention_mask position_idstorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096])batch tensor after cp: + position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor: tokens batch tensor after cp:torch.Size([4, 4096]) tokens + batch tensor:torch.Size([4, 4096]) +labels batch tensor after cp:torch.Size([4, 4096]) +labelsbatch tensor: torch.Size([4, 4096])loss_mask + torch.Size([4, 4096])batch tensor after cp: + batch tensor:loss_mask attention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor after cp:batch tensor: attention_maskposition_ids torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:20:03] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 1423.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:20:04] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 1091.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor: batch tensor:tokens tokens batch tensor after cp: tokenstorch.Size([4, 4096]) torch.Size([4, 4096])torch.Size([4, 4096]) + + +batch tensor after cp:batch tensor: batch tensor:labelslabels torch.Size([4, 4096])labelstorch.Size([4, 4096]) + +batch tensor after cp:torch.Size([4, 4096])batch tensor: + batch tensor:loss_maskloss_mask loss_mask torch.Size([4, 4096]) +torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor after cp: attention_maskbatch tensor:attention_mask attention_masktorch.Size([4, 1, 4096, 4096])torch.Size([4, 1, 4096, 4096]) + +torch.Size([4, 1, 4096, 4096])batch tensor:batch tensor after cp: + position_idsposition_idsbatch tensor: torch.Size([4, 4096])position_idstorch.Size([4, 4096]) + +torch.Size([4, 4096]) +batch tensor: tokens batch tensor after cp:batch tensor after cp:torch.Size([4, 4096]) tokenstokens + torch.Size([4, 4096])torch.Size([4, 4096])batch tensor: + + labelsbatch tensor after cp: batch tensor after cp: torch.Size([4, 4096]) labels + labelsbatch tensor: torch.Size([4, 4096]) torch.Size([4, 4096]) +loss_mask +batch tensor after cp: batch tensor after cp:loss_mask torch.Size([4, 4096]) loss_mask +torch.Size([4, 4096]) +batch tensor:torch.Size([4, 4096])batch tensor after cp: + batch tensor after cp: attention_mask attention_mask attention_mask torch.Size([4, 1, 4096, 4096])torch.Size([4, 1, 4096, 4096])torch.Size([4, 1, 4096, 4096]) + + +batch tensor after cp:batch tensor after cp:batch tensor: position_ids position_idsposition_ids torch.Size([4, 4096]) torch.Size([4, 4096]) +torch.Size([4, 4096]) + +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask batch tensor:torch.Size([4, 1, 4096, 4096]) +batch tensor: position_idstokens torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids batch tensor after cp:torch.Size([4, 4096]) +tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:20:04] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 49.1 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens batch tensor:torch.Size([4, 4096]) + batch tensor:tokens labels torch.Size([4, 4096]) +batch tensor:torch.Size([4, 4096]) +loss_mask batch tensor:torch.Size([4, 4096]) +labels batch tensor:torch.Size([4, 4096]) +attention_maskbatch tensor: loss_masktorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096]) +batch tensor: batch tensor:position_ids attention_mask torch.Size([4, 4096]) +torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096])batch tensor after cp: + batch tensor after cp:tokens labelstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:labels loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: + loss_maskbatch tensor after cp: torch.Size([4, 4096])attention_mask + batch tensor after cp:torch.Size([4, 1, 4096, 4096]) +attention_mask torch.Size([4, 1, 4096, 4096])batch tensor after cp: + position_idsbatch tensor after cp: torch.Size([4, 4096])position_ids + torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:20:04] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 41.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor:batch tensor after cp: tokens torch.Size([4, 4096]) +tokensbatch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp:torch.Size([4, 4096]) loss_mask + torch.Size([4, 4096]) +batch tensor:batch tensor after cp: labelsattention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor after cp: batch tensor:position_ids loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor:batch tensor: tokens tokens torch.Size([4, 4096]) +batch tensor:torch.Size([4, 4096]) batch tensor after cp:labels + torch.Size([4, 4096])tokens +batch tensor: batch tensor: torch.Size([4, 4096]) +labelsloss_mask batch tensor after cp: torch.Size([4, 4096])torch.Size([4, 4096])labels + + batch tensor:batch tensor:torch.Size([4, 4096]) attention_mask +loss_mask batch tensor after cp: torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096])loss_mask + + batch tensor:torch.Size([4, 4096]) batch tensor: +position_ids batch tensor after cp: attention_mask torch.Size([4, 4096]) +attention_masktorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 1, 4096, 4096])batch tensor: + batch tensor after cp:position_ids position_idstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:loss_mask tokenstorch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: + attention_maskbatch tensor after cp: labelstorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096])batch tensor after cp: + position_idsbatch tensor after cp: torch.Size([4, 4096])loss_mask + torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:20:04] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 40.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor after cp: batch tensor:tokens labelstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: batch tensor:labels loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: + loss_mask batch tensor:torch.Size([4, 4096]) +attention_mask batch tensor after cp:torch.Size([4, 1, 4096, 4096]) +attention_mask batch tensor:torch.Size([4, 1, 4096, 4096]) +position_ids batch tensor after cp:torch.Size([4, 4096]) +position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096])batch tensor: +batch tensor after cp: position_ids tokenstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids batch tensor:torch.Size([4, 4096]) + tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor:batch tensor after cp: loss_masktokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp: batch tensor:labels attention_masktorch.Size([4, 4096]) + batch tensor after cp:torch.Size([4, 1, 4096, 4096]) +loss_mask batch tensor:torch.Size([4, 4096]) +position_idsbatch tensor after cp: torch.Size([4, 4096])attention_mask + batch tensor:torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_idstokens torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor after cp: batch tensor:tokens loss_mask batch tensor:torch.Size([4, 4096])torch.Size([4, 4096]) + + batch tensor:batch tensor after cp: tokenslabelsattention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +torch.Size([4, 4096])batch tensor after cp:batch tensor: + loss_maskposition_idsbatch tensor: torch.Size([4, 4096])labelstorch.Size([4, 4096]) + +torch.Size([4, 4096]) +batch tensor after cp: batch tensor:attention_mask loss_mask torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor: position_ids attention_masktorch.Size([4, 4096]) +torch.Size([4, 1, 4096, 4096]) +batch tensor after cp:batch tensor: tokensposition_ids torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:20:04] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 40.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens batch tensor:torch.Size([4, 4096]) + batch tensor:tokens labels torch.Size([4, 4096]) +batch tensor: loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor:batch tensor: labelsattention_mask torch.Size([4, 4096]) +torch.Size([4, 1, 4096, 4096])batch tensor: + loss_maskbatch tensor: position_idstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096])batch tensor after cp: + batch tensor after cp:tokens loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:labels attention_masktorch.Size([4, 4096]) +torch.Size([4, 1, 4096, 4096])batch tensor after cp: + batch tensor after cp:loss_mask torch.Size([4, 4096])position_ids + torch.Size([4, 4096])batch tensor after cp: + attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096])batch tensor: +batch tensor after cp: tokenslabels torch.Size([4, 4096]) +batch tensor after cp: torch.Size([4, 4096])loss_mask +torch.Size([4, 4096]) +batch tensor:batch tensor after cp: labelsattention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor:batch tensor after cp: loss_maskposition_ids torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096])batch tensor: +batch tensor: position_ids batch tensor: tokenstorch.Size([4, 4096]) + tokens torch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor: labels + torch.Size([4, 4096]) +batch tensor:batch tensor: labelsbatch tensor after cp: loss_mask torch.Size([4, 4096]) tokens +torch.Size([4, 4096]) batch tensor: +torch.Size([4, 4096]) loss_mask +batch tensor:torch.Size([4, 4096])batch tensor after cp: + attention_masklabels batch tensor: torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096])attention_mask + + batch tensor after cp:batch tensor:torch.Size([4, 1, 4096, 4096]) +loss_maskposition_ids batch tensor: torch.Size([4, 4096]) torch.Size([4, 4096])position_ids + torch.Size([4, 4096]) + +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:tokens tokenstorch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: + labelsbatch tensor after cp: labelstorch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: + loss_maskbatch tensor after cp: torch.Size([4, 4096])loss_mask + batch tensor after cp:torch.Size([4, 4096]) +attention_maskbatch tensor after cp: attention_masktorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 1, 4096, 4096])batch tensor after cp: + position_idsbatch tensor after cp: torch.Size([4, 4096])position_ids + torch.Size([4, 4096]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:20:04] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 41.4 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens batch tensor:torch.Size([4, 4096]) +tokens batch tensor: labels torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor: loss_maskbatch tensor: torch.Size([4, 4096])labels + torch.Size([4, 4096])batch tensor: + attention_maskbatch tensor: loss_masktorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096]) +batch tensor: position_idsbatch tensor: torch.Size([4, 4096])attention_mask + torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp:batch tensor after cp: loss_masktokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: labelsattention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor after cp:batch tensor after cp: loss_maskposition_ids torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor:batch tensor:batch tensor: labels tokens tokens torch.Size([4, 4096]) +batch tensor: torch.Size([4, 4096])loss_mask +torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor: batch tensor:attention_mask labels labels torch.Size([4, 1, 4096, 4096]) torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: + position_idsbatch tensor: batch tensor: torch.Size([4, 4096]) loss_mask +loss_mask torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor: attention_maskattention_mask torch.Size([4, 1, 4096, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor:batch tensor: position_idsposition_ids torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask batch tensor after cp:torch.Size([4, 1, 4096, 4096]) batch tensor after cp: +tokens tokensbatch tensor after cp: torch.Size([4, 4096])torch.Size([4, 4096]) + +position_idsbatch tensor after cp: batch tensor after cp: torch.Size([4, 4096]) labels +labels torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: loss_maskloss_mask torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: attention_maskattention_mask torch.Size([4, 1, 4096, 4096]) +torch.Size([4, 1, 4096, 4096])batch tensor after cp: + batch tensor after cp:position_ids position_idstorch.Size([4, 4096]) +torch.Size([4, 4096]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:20:04] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 41.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens batch tensor: torch.Size([4, 4096])tokens + batch tensor: labels torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor: loss_maskbatch tensor: labelstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor:batch tensor: attention_maskloss_mask torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096]) + +batch tensor: batch tensor:position_ids attention_mask torch.Size([4, 4096]) +torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids batch tensor:torch.Size([4, 4096]) + tokens torch.Size([4, 4096]) +batch tensor after cp:batch tensor: tokenslabels torch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: +batch tensor after cp: tokenslabels batch tensor:torch.Size([4, 4096]) +loss_masktorch.Size([4, 4096]) batch tensor after cp: +torch.Size([4, 4096]) batch tensor after cp: +labels loss_maskbatch tensor: torch.Size([4, 4096]) +torch.Size([4, 4096])attention_maskbatch tensor after cp: + batch tensor after cp:torch.Size([4, 1, 4096, 4096]) loss_mask +attention_mask batch tensor:torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +position_idsbatch tensor after cp: batch tensor after cp: torch.Size([4, 4096]) position_idsattention_mask + torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096]) + +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor:batch tensor after cp: labels torch.Size([4, 4096])tokens + batch tensor after cp: loss_mask torch.Size([4, 4096]) +torch.Size([4, 4096])batch tensor after cp: + attention_mask batch tensor:torch.Size([4, 1, 4096, 4096]) +labelsbatch tensor after cp: position_idstorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp:batch tensor: tokens torch.Size([4, 4096]) +tokensbatch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: torch.Size([4, 4096])loss_mask +torch.Size([4, 4096]) +batch tensor:batch tensor after cp: labelsattention_mask torch.Size([4, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor:batch tensor after cp: loss_maskposition_ids torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids batch tensor:torch.Size([4, 4096]) + tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: batch tensor:attention_mask torch.Size([4, 1, 4096, 4096]) + batch tensor after cp:tokens position_ids torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:20:04] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 39.9 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:20:04 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.024567365646362305 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.024675846099853516 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.024802684783935547 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.02576160430908203 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.02578282356262207 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.02581334114074707 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.025858640670776367 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.02772378921508789 to prepare state dict for ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.517096757888794 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.517061710357666 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.5171067714691162 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.5165226459503174 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.5171444416046143 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.5184874534606934 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 0.009727954864501953 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.5196855068206787 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, plan time: 0.013611078262329102 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, plan time: 0.013528823852539062 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, plan time: 0.013626813888549805 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, plan time: 0.01360464096069336 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, plan time: 0.013625383377075195 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3048382 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3048382 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3048384 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3048418 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.304845 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.059906005859375e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.489059448242188e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 8.96453857421875e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.5367431640625e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.608268737792969e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, plan time: 0.009649991989135742 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3051643 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, plan time: 0.0034627914428710938 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3052123 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.0001285076141357422 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00016450881958007812 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, plan time: 0.01553201675415039 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540807.3098202 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010657310485839844 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04439806938171387 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.349696 rank: 6, write(async) time: 0.04485487937927246 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04471230506896973 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.349983 rank: 5, write(async) time: 0.04514336585998535 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.044977426528930664 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.045075416564941406 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.3502362 rank: 3, write(async) time: 0.04539084434509277 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04517316818237305 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.350348 rank: 7, write(async) time: 0.04550480842590332 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.3504658 rank: 2, write(async) time: 0.04562950134277344 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04453158378601074 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.3547986 rank: 0, write(async) time: 0.04497933387756348 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.06074714660644531 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.3664086 rank: 1, write(async) time: 0.061241865158081055 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.0726020336151123 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540807.3783205 rank: 4, write(async) time: 0.07310748100280762 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 1.52587890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 1.4543533325195312e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 1.6689300537109375e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 1.6927719116210938e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 1.5497207641601562e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 2.3126602172851562e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 1.4543533325195312e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 0.024932861328125 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 0.027431011199951172 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 0.025368928909301758 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 0.025101900100708008 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 0.025771617889404297 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 0.027016639709472656 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 0.036684513092041016 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 1.4543533325195312e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 0.03063368797302246 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214347776, before: 1630994432, after: 1845342208 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214470656, before: 1618747392, after: 1833218048 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214327296, before: 1617031168, after: 1831358464 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214433792, before: 1611845632, after: 1826279424 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214384640, before: 1614974976, after: 1829359616 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216326144, before: 1614979072, after: 1831305216 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214441984, before: 1630273536, after: 1844715520 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216522752, before: 1617035264, after: 1833558016 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216424448, before: 1618747392, after: 1835171840 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.1467986, rank: 6, write(sync,parallel): 0.582798957824707 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216608768, before: 1611849728, after: 1828458496 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 215863296, before: 1630277632, after: 1846140928 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216522752, before: 1630961664, after: 1847484416 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.1736574, rank: 1, write(sync,parallel): 0.6021838188171387 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216375296, before: 1655816192, after: 1872191488 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214347776, before: 1655889920, after: 1870237696 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.65s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.1803813, rank: 5, write(sync,parallel): 0.6167054176330566 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.1889162, rank: 3, write(sync,parallel): 0.6200227737426758 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.2035947, rank: 4, write(sync,parallel): 0.6129310131072998 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.67s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.2168362, rank: 7, write(sync,parallel): 0.6513214111328125 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 227573760, before: 1900617728, after: 2128191488 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 229113856, before: 1900617728, after: 2129731584 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.2391086, rank: 2, write(sync,parallel): 0.669452428817749 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.70s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.72s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540808.265447, rank: 0, write(sync,parallel): 0.6024320125579834 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.74s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.67s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.30342, 2, gather: 0.023463726043701172 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.303432, 1, gather: 0.09185409545898438 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.3035126, 4, gather: 0.05437421798706055 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.3035476, 7, gather: 0.05009913444519043 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.3035488, 5, gather: 0.08676028251647949 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.303551, 3, gather: 0.07970476150512695 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.303572, 6, gather: 0.12238788604736328 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.305068, 0, gather: 0.003711700439453125 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540808.3231246, metadata_write: 0.017920255661010742 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0454s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0239s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0717s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1440s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1013s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1084s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0761s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1138s + successfully saved checkpoint from iteration 10 to gpt-checkpoint [ t 1/8, p 1/1 ] +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.0024979114532470703 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.002519369125366211 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.002494335174560547 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.002518177032470703 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.0025148391723632812 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.0025131702423095703 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.0025305747985839844 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.0025055408477783203 to finalize ckpt save +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor:batch tensor: tokenstokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor: labelsbatch tensor: torch.Size([4, 4096])labels + batch tensor:torch.Size([4, 4096]) +loss_maskbatch tensor: torch.Size([4, 4096])loss_mask + torch.Size([4, 4096])batch tensor: + attention_maskbatch tensor: torch.Size([4, 1, 4096, 4096]) +attention_mask batch tensor:torch.Size([4, 1, 4096, 4096]) +position_ids batch tensor:torch.Size([4, 4096]) +position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp:batch tensor after cp: loss_masktokens torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor after cp: batch tensor after cp:attention_mask labels torch.Size([4, 1, 4096, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: position_idsloss_mask torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor after cp: attention_mask tokenstorch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: batch tensor:labels torch.Size([4, 4096]) +tokensbatch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: torch.Size([4, 4096])attention_mask +torch.Size([4, 1, 4096, 4096]) +batch tensor:batch tensor: labelsposition_ids torch.Size([4, 4096]) + batch tensor: torch.Size([4, 4096])loss_mask + torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids batch tensor after cp:torch.Size([4, 4096]) +tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 10 +Done exporting trace 10 +(min, max) time across ranks (ms): + evaluate .......................................: (1492.92, 1497.73) +---------------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on validation set | lm loss value: 1.178511E+01 | lm loss PPL: 1.312835E+05 | +---------------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor:batch tensor: tokenstokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor: labelslabels torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor: loss_maskloss_mask torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor:batch tensor: attention_maskattention_mask torch.Size([4, 1, 4096, 4096])torch.Size([4, 1, 4096, 4096]) + +batch tensor:batch tensor: position_idsposition_ids torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: tokenstokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor after cp: labelslabels torch.Size([4, 4096])batch tensor:torch.Size([4, 4096]) + + batch tensor after cp: batch tensor after cp:loss_masktokens loss_masktorch.Size([4, 4096]) + torch.Size([4, 4096])torch.Size([4, 4096])batch tensor after cp: + + batch tensor after cp:attention_maskbatch tensor: torch.Size([4, 1, 4096, 4096]) attention_mask +labels batch tensor after cp: torch.Size([4, 1, 4096, 4096]) torch.Size([4, 4096]) +position_ids +batch tensor after cp: batch tensor:position_idstorch.Size([4, 4096]) + loss_masktorch.Size([4, 4096]) +torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor:batch tensor after cp: labelstokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor: labelsloss_mask torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp:batch tensor: loss_maskattention_mask torch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096])batch tensor: + position_idsbatch tensor after cp: torch.Size([4, 4096])attention_mask + torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokensbatch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp:torch.Size([4, 4096]) labels + torch.Size([4, 4096]) +batch tensor:batch tensor after cp: labelsloss_mask torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp: batch tensor:attention_mask loss_masktorch.Size([4, 1, 4096, 4096]) +torch.Size([4, 4096])batch tensor after cp: +batch tensor: batch tensor:position_ids torch.Size([4, 4096])attention_masktokens + torch.Size([4, 1, 4096, 4096]) +batch tensor: position_idstorch.Size([4, 4096]) torch.Size([4, 4096]) + +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask batch tensor after cp:torch.Size([4, 1, 4096, 4096]) +tokens batch tensor: torch.Size([4, 4096])position_ids + batch tensor after cp:torch.Size([4, 4096]) +labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: batch tensor after cp:position_ids tokens torch.Size([4, 4096])torch.Size([4, 4096]) + +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +batch tensor: tokens torch.Size([4, 4096]) +batch tensor: labels torch.Size([4, 4096]) +batch tensor: loss_mask torch.Size([4, 4096]) +batch tensor: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor: position_ids torch.Size([4, 4096]) +batch tensor after cp: tokens torch.Size([4, 4096]) +batch tensor after cp: labels torch.Size([4, 4096]) +batch tensor after cp: loss_mask torch.Size([4, 4096]) +batch tensor after cp: attention_mask torch.Size([4, 1, 4096, 4096]) +batch tensor after cp: position_ids torch.Size([4, 4096]) +Start exporting trace 11 +Done exporting trace 11 +(min, max) time across ranks (ms): + evaluate .......................................: (19.82, 23.28) +---------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on test set | lm loss value: 1.178511E+01 | lm loss PPL: 1.312835E+05 | +---------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Running ctx_length=2048, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 2048 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 2048 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 2048 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 2048 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +> compiling dataset index builder ... +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.049 seconds +> compiling and loading fused kernels ... +INFO:megatron.training.initialize:Setting logging level to 0 +>>> done with compiling and loading fused kernels. Compilation time: 2.568 seconds +time to initialize megatron (seconds): 7.576 +[after megatron is initialized] datetime: 2025-06-21 21:20:47 +building GPT model ... +>>> embedding>>> embedding + +>>> decoder>>> decoder + +>>> output_layer>>> output_layer + +>>> embedding +>>> decoder +>>> output_layer +>>> embedding>>> embedding + +>>> decoder +>>> decoder>>> output_layer + +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 78706176 > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 78706176 + + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 78706176 + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 78706176 > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 78706176 + + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 78706176 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 78706176 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 78706176 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (78706176 elements, 78706176 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.embedding.position_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.final_layernorm.bias + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.0.self_attention.linear_proj.bias +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (3.31, 3.33) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:20:47 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=2048, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.005295 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 33296 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.002420 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 33281 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.002396 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 33343 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:20:48 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (296.82, 329.76) + train/valid/test-data-iterators-setup ..........: (29.51, 115.28) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:20:48 +batch tensor: tokens batch tensor: torch.Size([4, 8192])tokens +batch tensor: labels torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: batch tensor:labels attention_mask torch.Size([4, 8192]) +torch.Size([4, 1, 8192, 8192]) +batch tensor: batch tensor:loss_mask position_idstorch.Size([4, 8192]) +torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp:batch tensor after cp: position_idstokens torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:20:55] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 7793.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 2] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6314.0 | max reserved: 6314.0 +[Rank 3] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6314.0 | max reserved: 6314.0 +[Rank 7] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6314.0 | max reserved: 6314.0 +[Rank 5] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6826.0 | max reserved: 6826.0 +[Rank 4] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6314.0 | max reserved: 6314.0 +[Rank 0] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6314.0 | max reserved: 6314.0 +[Rank 6] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6058.0 | max reserved: 6058.0[Rank 1] (after 1 iterations) memory (MB) | allocated: 1295.60107421875 | max allocated: 5998.04931640625 | reserved: 6058.0 | max reserved: 6058.0 + +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask batch tensor:torch.Size([4, 1, 8192, 8192]) + batch tensor: tokensposition_ids torch.Size([4, 8192]) +torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_maskbatch tensor: torch.Size([4, 8192]) +tokensbatch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: torch.Size([4, 8192])position_ids +torch.Size([4, 8192])batch tensor: + labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:20:56] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 136.9 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor:batch tensor after cp: labelstokens torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor:batch tensor after cp: loss_masklabels torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor: loss_mask attention_masktorch.Size([4, 8192]) +torch.Size([4, 1, 8192, 8192])batch tensor after cp: + attention_maskbatch tensor: torch.Size([4, 1, 8192, 8192])position_ids + batch tensor after cp:torch.Size([4, 8192]) +position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:20:56] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 101.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192])batch tensor: +batch tensor: tokens labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor:batch tensor: attention_masklabels torch.Size([4, 1, 8192, 8192])torch.Size([4, 8192]) + +batch tensor:batch tensor: position_idsloss_mask torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: batch tensor after cp:tokens attention_masktorch.Size([4, 8192]) +torch.Size([4, 1, 8192, 8192])batch tensor after cp: + labelsbatch tensor after cp: torch.Size([4, 8192])position_ids + batch tensor after cp:torch.Size([4, 8192]) +loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:20:56] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 93.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor:batch tensor after cp: tokens torch.Size([4, 8192])tokens + batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp:torch.Size([4, 8192]) +loss_mask batch tensor:torch.Size([4, 8192]) +labels batch tensor after cp: torch.Size([4, 8192])attention_mask + batch tensor:torch.Size([4, 1, 8192, 8192]) +loss_maskbatch tensor after cp: torch.Size([4, 8192])position_ids + batch tensor:torch.Size([4, 8192]) +attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor:batch tensor: tokens tokens torch.Size([4, 8192]) +torch.Size([4, 8192])batch tensor: + labels batch tensor:torch.Size([4, 8192]) +labelsbatch tensor: torch.Size([4, 8192])loss_mask + batch tensor:torch.Size([4, 8192]) +loss_mask batch tensor:torch.Size([4, 8192]) +attention_mask batch tensor:torch.Size([4, 1, 8192, 8192]) +attention_maskbatch tensor: position_idstorch.Size([4, 1, 8192, 8192]) +torch.Size([4, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp:batch tensor after cp: tokenstokens torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor after cp: labelslabels torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor after cp: loss_maskloss_mask torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor after cp: attention_maskattention_mask torch.Size([4, 1, 8192, 8192]) +torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: batch tensor after cp:position_ids position_idstorch.Size([4, 8192]) +torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:20:56] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 88.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:20:56] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 93.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor:batch tensor after cp: tokens torch.Size([4, 8192])tokens + batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp:torch.Size([4, 8192]) loss_mask + torch.Size([4, 8192])batch tensor: + batch tensor after cp:labels torch.Size([4, 8192])attention_mask + batch tensor:torch.Size([4, 1, 8192, 8192]) + batch tensor after cp:loss_mask position_idstorch.Size([4, 8192]) +torch.Size([4, 8192])batch tensor: + attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_maskbatch tensor: torch.Size([4, 1, 8192, 8192]) +tokens batch tensor: position_ids torch.Size([4, 8192]) +torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192])batch tensor after cp: + tokensbatch tensor: torch.Size([4, 8192])position_ids + torch.Size([4, 8192])batch tensor after cp: + labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:20:56] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 88.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:20:56] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 88.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor:batch tensor: tokens tokens torch.Size([4, 8192]) +torch.Size([4, 8192])batch tensor: + labels batch tensor:torch.Size([4, 8192]) +labelsbatch tensor: torch.Size([4, 8192])loss_mask + batch tensor:torch.Size([4, 8192]) +loss_mask batch tensor:torch.Size([4, 8192]) +attention_mask batch tensor: torch.Size([4, 1, 8192, 8192])attention_mask + batch tensor:torch.Size([4, 1, 8192, 8192]) +position_ids batch tensor:torch.Size([4, 8192]) +position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens batch tensor after cp:torch.Size([4, 8192]) +tokensbatch tensor after cp: torch.Size([4, 8192])labels + batch tensor after cp:torch.Size([4, 8192]) + batch tensor after cp:labels loss_mask torch.Size([4, 8192]) +torch.Size([4, 8192]) +batch tensor after cp: batch tensor after cp:loss_mask attention_mask torch.Size([4, 8192])torch.Size([4, 1, 8192, 8192]) + +batch tensor after cp:batch tensor after cp: attention_maskposition_ids torch.Size([4, 1, 8192, 8192])torch.Size([4, 8192]) + +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor:batch tensor: labels torch.Size([4, 8192])tokens + batch tensor: loss_mask torch.Size([4, 8192]) +torch.Size([4, 8192])batch tensor: + attention_mask batch tensor:torch.Size([4, 1, 8192, 8192]) +labels batch tensor:torch.Size([4, 8192]) +position_idsbatch tensor: torch.Size([4, 8192])loss_mask + torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp:batch tensor after cp: attention_masktokens torch.Size([4, 1, 8192, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor after cp: position_idslabels torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:20:56] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 85.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:20:56] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 87.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:20:56 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.04579305648803711 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.045844078063964844 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.04585766792297363 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.04589247703552246 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.0459589958190918 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.04596996307373047 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.04594874382019043 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.04618215560913086 to prepare state dict for ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0734870433807373 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0734751224517822 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0735414028167725 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0738377571105957 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0737571716308594 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0740272998809814 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.0740764141082764 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 0.007040739059448242 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, plan time: 0.006707906723022461 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, plan time: 0.00654149055480957 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, plan time: 0.0065953731536865234 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, plan time: 0.006337404251098633 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, plan time: 0.006575584411621094 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.902314 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.902315 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.9023194 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.902321 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.9023414 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, plan time: 0.006052970886230469 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, plan time: 0.00613713264465332 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.9023945 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010013580322265625 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.894371032714844e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.965896606445312e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010657310485839844 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.0001087188720703125 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.9024434 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00011730194091796875 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.655952453613281e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, plan time: 0.0075418949127197266 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540858.9072328 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010633468627929688 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.03879880905151367 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9415622 rank: 6, write(async) time: 0.039246559143066406 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.03901505470275879 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.039075374603271484 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9417899 rank: 2, write(async) time: 0.039469242095947266 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9418366 rank: 1, write(async) time: 0.0395197868347168 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.03950619697570801 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9422946 rank: 4, write(async) time: 0.03995084762573242 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.03950357437133789 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.947155 rank: 0, write(async) time: 0.03991985321044922 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.046088457107543945 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9488823 rank: 5, write(async) time: 0.04655909538269043 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04618954658508301 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9490929 rank: 3, write(async) time: 0.04664778709411621 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.048691749572753906 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540858.9515853 rank: 7, write(async) time: 0.04918980598449707 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 1.430511474609375e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 1.52587890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 1.5735626220703125e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 1.621246337890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 1.4543533325195312e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 1.5020370483398438e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 1.430511474609375e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 0.02055644989013672 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 0.02059793472290039 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 0.0204775333404541 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 0.023104190826416016 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 0.026561260223388672 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 0.026290178298950195 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 0.028755664825439453 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 1.430511474609375e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 0.024779558181762695 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216412160, before: 1656692736, after: 1873104896 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214532096, before: 1654870016, after: 1869402112 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214401024, before: 1656692736, after: 1871093760 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214458368, before: 1662509056, after: 1876967424 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.696301, rank: 2, write(sync,parallel): 0.5541393756866455 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.61s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 215896064, before: 1648365568, after: 1864261632 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 215592960, before: 1662500864, after: 1878093824 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214536192, before: 1648365568, after: 1862901760 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214401024, before: 1642954752, after: 1857355776 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214327296, before: 1686175744, after: 1900503040 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216580096, before: 1642954752, after: 1859534848 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.820496, rank: 6, write(sync,parallel): 0.67242431640625 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216023040, before: 1686175744, after: 1902198784 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214470656, before: 1671000064, after: 1885470720 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216424448, before: 1671000064, after: 1887424512 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216674304, before: 1654870016, after: 1871544320 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.8306322, rank: 1, write(sync,parallel): 0.6838984489440918 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.72s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.74s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.861314, rank: 5, write(sync,parallel): 0.7025630474090576 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.8637755, rank: 4, write(sync,parallel): 0.7129194736480713 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.868855, rank: 3, write(sync,parallel): 0.7087769508361816 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.8887198, rank: 7, write(sync,parallel): 0.7259976863861084 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.77s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.77s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 241725440, before: 1915785216, after: 2157510656 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.78s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.79s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 240295936, before: 1915785216, after: 2156081152 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540859.9657967, rank: 0, write(sync,parallel): 0.711611270904541 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.77s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.997716, 1, gather: 0.13897466659545898 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.9978263, 2, gather: 0.2699146270751953 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.9979436, 4, gather: 0.10540199279785156 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.9979637, 3, gather: 0.09132933616638184 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.998051, 5, gather: 0.10126566886901855 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.9980783, 6, gather: 0.14963221549987793 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.9980984, 7, gather: 0.07274627685546875 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540859.9999723, 0, gather: 0.0042150020599365234 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540860.0104012, metadata_write: 0.010300636291503906 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1198s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1156s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1638s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0170s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.2843s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1057s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0871s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.1536s + successfully saved checkpoint from iteration 10 to gpt-checkpoint [ t 1/8, p 1/1 ] +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.0023696422576904297 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.0023462772369384766 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.002360820770263672 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.002358675003051758 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.002367258071899414 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.002351999282836914 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.002353668212890625 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.002376079559326172 to finalize ckpt save +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_maskbatch tensor: torch.Size([4, 8192]) + batch tensor after cp: tokensattention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 10 +Done exporting trace 10 +(min, max) time across ranks (ms): + evaluate .......................................: (1406.99, 1408.20) +---------------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on validation set | lm loss value: 1.138065E+01 | lm loss PPL: 8.761001E+04 | +---------------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor:batch tensor: tokenstokens torch.Size([4, 8192]) +torch.Size([4, 8192])batch tensor: +labels torch.Size([4, 8192]) +batch tensor: batch tensor:labels loss_maskbatch tensor: torch.Size([4, 8192])torch.Size([4, 8192]) + +tokensbatch tensor: batch tensor: loss_masktorch.Size([4, 8192])attention_mask + torch.Size([4, 8192]) +torch.Size([4, 1, 8192, 8192])batch tensor: + labelsbatch tensor: batch tensor: torch.Size([4, 8192]) position_ids + attention_masktorch.Size([4, 8192])batch tensor: + loss_masktorch.Size([4, 1, 8192, 8192]) +torch.Size([4, 8192])batch tensor: + position_idsbatch tensor: torch.Size([4, 8192])attention_mask + torch.Size([4, 1, 8192, 8192])batch tensor after cp: + tokensbatch tensor: torch.Size([4, 8192])position_ids +torch.Size([4, 8192])batch tensor after cp: + labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp:batch tensor after cp: tokenslabels torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor after cp: labelsloss_mask torch.Size([4, 8192])torch.Size([4, 8192]) + +batch tensor after cp:batch tensor after cp: loss_maskattention_mask torch.Size([4, 8192]) +torch.Size([4, 1, 8192, 8192])batch tensor after cp: + batch tensor after cp:attention_mask position_ids torch.Size([4, 1, 8192, 8192])torch.Size([4, 8192]) + +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +batch tensor: tokens torch.Size([4, 8192]) +batch tensor: labels torch.Size([4, 8192]) +batch tensor: loss_mask torch.Size([4, 8192]) +batch tensor: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor: position_ids torch.Size([4, 8192]) +batch tensor after cp: tokens torch.Size([4, 8192]) +batch tensor after cp: labels torch.Size([4, 8192]) +batch tensor after cp: loss_mask torch.Size([4, 8192]) +batch tensor after cp: attention_mask torch.Size([4, 1, 8192, 8192]) +batch tensor after cp: position_ids torch.Size([4, 8192]) +Start exporting trace 11 +Done exporting trace 11 +(min, max) time across ranks (ms): + evaluate .......................................: (46.94, 47.97) +---------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on test set | lm loss value: 1.138065E+01 | lm loss PPL: 8.761001E+04 | +---------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Running ctx_length=4096, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 4096 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 4096 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 4096 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 4096 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.043 seconds +> compiling and loading fused kernels ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +>>> done with compiling and loading fused kernels. Compilation time: 2.576 seconds +time to initialize megatron (seconds): 7.565 +[after megatron is initialized] datetime: 2025-06-21 21:21:39 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 87094784 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 87094784 + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 87094784 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 87094784 + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 87094784 + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 87094784 + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 87094784 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 87094784 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (87094784 elements, 87094784 padded size): + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.final_layernorm.weight + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.embedding.position_embeddings.weight + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.final_layernorm.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_proj.bias +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (3.29, 3.36) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:21:39 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.006160 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 16648 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.002159 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 16640 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.002049 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 16671 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:21:40 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (522.66, 551.25) + train/valid/test-data-iterators-setup ..........: (100.28, 182.78) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:21:40 +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:21:47] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 7488.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 5] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 13610.0 | max reserved: 13610.0 +[Rank 4] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12074.0 | max reserved: 12074.0[Rank 6] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12842.0 | max reserved: 12842.0 + +[Rank 3] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12586.0 | max reserved: 12586.0 +[Rank 7] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12842.0 | max reserved: 12842.0 +[Rank 1] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12842.0 | max reserved: 12842.0 +[Rank 2] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12074.0 | max reserved: 12074.0 +[Rank 0] (after 1 iterations) memory (MB) | allocated: 2160.47607421875 | max allocated: 11907.61181640625 | reserved: 12842.0 | max reserved: 12842.0 +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labelsbatch tensor after cp: torch.Size([4, 16384])tokens + batch tensor:torch.Size([4, 16384]) +loss_mask batch tensor after cp:torch.Size([4, 16384]) +labels batch tensor: torch.Size([4, 16384])attention_mask + batch tensor after cp:torch.Size([4, 1, 16384, 16384]) +loss_maskbatch tensor: torch.Size([4, 16384])position_ids + batch tensor after cp:torch.Size([4, 16384]) +attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:21:47] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 294.9 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:21:48] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 265.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:21:48] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 257.3 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:21:48] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 270.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor:batch tensor: labels tokenstorch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +torch.Size([4, 16384])batch tensor: +attention_mask torch.Size([4, 1, 16384, 16384])batch tensor: + labelsbatch tensor: torch.Size([4, 16384])position_ids + batch tensor:torch.Size([4, 16384]) +loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:21:48] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 260.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:21:49] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 253.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor:batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +tokensbatch tensor after cp: position_ids torch.Size([4, 16384]) +torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:21:49] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 247.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens batch tensor:torch.Size([4, 16384]) + tokens batch tensor: labels torch.Size([4, 16384]) +batch tensor: torch.Size([4, 16384])loss_mask +torch.Size([4, 16384]) +batch tensor:batch tensor: labelsattention_mask torch.Size([4, 16384]) +torch.Size([4, 1, 16384, 16384])batch tensor: + loss_maskbatch tensor: position_idstorch.Size([4, 16384]) +torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp:batch tensor after cp: labelstokens torch.Size([4, 16384])torch.Size([4, 16384]) + +batch tensor after cp:batch tensor after cp: labelsloss_mask torch.Size([4, 16384]) +torch.Size([4, 16384])batch tensor after cp: + loss_maskbatch tensor after cp: torch.Size([4, 16384])attention_mask + batch tensor after cp:torch.Size([4, 1, 16384, 16384]) +attention_mask batch tensor after cp:torch.Size([4, 1, 16384, 16384]) +position_idsbatch tensor after cp: position_idstorch.Size([4, 16384]) +torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:21:49] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 256.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +batch tensor: tokens torch.Size([4, 16384]) +batch tensor: labels torch.Size([4, 16384]) +batch tensor: loss_mask torch.Size([4, 16384]) +batch tensor: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor: position_ids torch.Size([4, 16384]) +batch tensor after cp: tokens torch.Size([4, 16384]) +batch tensor after cp: labels torch.Size([4, 16384]) +batch tensor after cp: loss_mask torch.Size([4, 16384]) +batch tensor after cp: attention_mask torch.Size([4, 1, 16384, 16384]) +batch tensor after cp: position_ids torch.Size([4, 16384]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:21:49] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 270.1 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:21:49 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.0335850715637207 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.03364253044128418 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.03365015983581543 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.033718109130859375 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.03370499610900879 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.03372073173522949 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.034130096435546875 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.03917837142944336 to prepare state dict for ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +Running ctx_length=8192, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 8192 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 8192 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 8192 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 8192 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +INFO:megatron.training.initialize:Setting logging level to 0 +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +INFO:megatron.training.initialize:Setting logging level to 0 +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.045 seconds +> compiling and loading fused kernels ... +>>> done with compiling and loading fused kernels. Compilation time: 2.535 seconds +time to initialize megatron (seconds): 7.108 +[after megatron is initialized] datetime: 2025-06-21 21:22:59 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 103872000 +>>> embedding>>> embedding + +>>> decoder +>>> decoder>>> output_layer + +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 103872000 + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 103872000 +>>> embedding > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 103872000 + +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 103872000 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 103872000 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 103872000 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 103872000 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (103872000 elements, 103872000 padded size): + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.final_layernorm.bias + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.final_layernorm.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.embedding.position_embeddings.weight + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (22.20, 22.24) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:23:00 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=8192, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.004859 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 8324 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001823 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 8320 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001820 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 8335 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:23:00 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (538.66, 560.52) + train/valid/test-data-iterators-setup ..........: (26.91, 117.95) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:23:00 +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:23:07] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 7354.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 1] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27858.0 | max reserved: 27858.0[Rank 3] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27890.0 | max reserved: 27890.0 + +[Rank 2] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27890.0 | max reserved: 27890.0[Rank 6] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 28530.0 | max reserved: 28530.0[Rank 4] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27506.0 | max reserved: 27506.0 + + +[Rank 7] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27858.0 | max reserved: 27858.0[Rank 5] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27858.0 | max reserved: 27858.0 + +[Rank 0] (after 1 iterations) memory (MB) | allocated: 5426.22607421875 | max allocated: 25262.73681640625 | reserved: 27858.0 | max reserved: 27858.0 +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:23:08] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 999.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:23:09] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 966.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:23:10] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 989.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:23:11] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 972.3 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor: tokens batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: torch.Size([4, 32768])labels + torch.Size([4, 32768]) +batch tensor:batch tensor after cp: labelsloss_mask torch.Size([4, 32768])torch.Size([4, 32768]) + +batch tensor after cp:batch tensor: loss_maskattention_mask torch.Size([4, 1, 32768, 32768])torch.Size([4, 32768]) + +batch tensor after cp: batch tensor:position_ids attention_masktorch.Size([4, 32768]) +torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:23:12] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 978.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:23:13] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 976.4 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:23:14] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 983.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:23:15] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 985.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:23:16] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 980.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:23:16 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.03102421760559082 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.031038284301757812 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.031049251556396484 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.0310518741607666 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.031064987182617188 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.031113862991333008 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.031130075454711914 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.03212904930114746 to prepare state dict for ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.2768168449401855 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.2768735885620117 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.2768831253051758 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.2768712043762207 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.277141809463501 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.2772819995880127 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.2767083644866943 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 0.007138967514038086 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, plan time: 0.005838155746459961 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, plan time: 0.0057871341705322266 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, plan time: 0.005843162536621094 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, plan time: 0.0058193206787109375 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427283 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, plan time: 0.005429983139038086 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427316 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427328 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427333 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, plan time: 0.005129575729370117 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, plan time: 0.004840373992919922 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427667 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427838 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9427786 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.441375732421875e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.036064147949219e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.369850158691406e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.72747802734375e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010561943054199219 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010275840759277344 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00011396408081054688 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, plan time: 0.007177591323852539 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750540998.9476008 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010061264038085938 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04657936096191406 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04671001434326172 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.046715497970581055 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.9898148 rank: 5, write(async) time: 0.04703712463378906 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.989935 rank: 6, write(async) time: 0.04720735549926758 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.989938 rank: 2, write(async) time: 0.04720592498779297 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.0488743782043457 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.992125 rank: 4, write(async) time: 0.049388885498046875 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.05014514923095703 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.05012106895446777 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.9933772 rank: 7, write(async) time: 0.05064249038696289 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.9934003 rank: 1, write(async) time: 0.05063295364379883 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.0504457950592041 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.993733 rank: 3, write(async) time: 0.0509490966796875 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.047002315521240234 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750540998.9950812 rank: 0, write(async) time: 0.04748106002807617 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 1.4543533325195312e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 1.4781951904296875e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 1.5497207641601562e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 1.5974044799804688e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 1.5497207641601562e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 1.8835067749023438e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 1.7404556274414062e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 0.027589082717895508 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 0.02630329132080078 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 0.02779674530029297 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 0.026431560516357422 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 0.02574324607849121 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 0.028816699981689453 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 0.027830123901367188 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 1.4066696166992188e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 0.030241012573242188 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216506368, before: 1636093952, after: 1852600320 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216662016, before: 1613303808, after: 1829965824 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214482944, before: 1612128256, after: 1826611200 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214425600, before: 1616084992, after: 1830510592 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216485888, before: 1612128256, after: 1828614144 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214511616, before: 1615822848, after: 1830334464 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214560768, before: 1646333952, after: 1860894720 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216719360, before: 1625911296, after: 1842630656 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.7614288, rank: 2, write(sync,parallel): 0.5582695007324219 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214708224, before: 1636093952, after: 1850802176 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216637440, before: 1615814656, after: 1832452096 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214515712, before: 1613303808, after: 1827819520 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 215719936, before: 1616084992, after: 1831804928 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214695936, before: 1625911296, after: 1840607232 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216559616, before: 1646333952, after: 1862893568 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.62s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.8080714, rank: 5, write(sync,parallel): 0.6000552177429199 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.819847, rank: 1, write(sync,parallel): 0.6087126731872559 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.8302224, rank: 3, write(sync,parallel): 0.6190035343170166 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.8311844, rank: 6, write(sync,parallel): 0.6246709823608398 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.83296, rank: 4, write(sync,parallel): 0.620896577835083 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750540999.8336303, rank: 7, write(sync,parallel): 0.6271798610687256 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.67s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.69s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.69s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.69s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 316727296, before: 1884409856, after: 2201137152 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 315887616, before: 1884409856, after: 2200297472 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541000.1216643, rank: 0, write(sync,parallel): 0.7498927116394043 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.82s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1601615, 4, gather: 0.2938997745513916 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1601605, 6, gather: 0.2925267219543457 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1602237, 5, gather: 0.312274694442749 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.160248, 7, gather: 0.2906785011291504 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1601899, 1, gather: 0.2974429130554199 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1601727, 2, gather: 0.3645768165588379 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1602547, 3, gather: 0.2923240661621094 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1624453, 0, gather: 0.004957675933837891 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541000.1776125, metadata_write: 0.015031814575195312 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3120s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3134s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3841s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0223s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3119s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3317s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3103s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.3173s + successfully saved checkpoint from iteration 10 to gpt-checkpoint [ t 1/8, p 1/1 ] +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.002425670623779297 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.0024383068084716797 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.0024361610412597656 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.002391815185546875 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.002431154251098633 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.0023288726806640625 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.0024106502532958984 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.002340555191040039 to finalize ckpt save +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 10 +Done exporting trace 10 +(min, max) time across ranks (ms): + evaluate .......................................: (2003.21, 2003.27) +---------------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on validation set | lm loss value: 1.358201E+01 | lm loss PPL: 7.917604E+05 | +---------------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +batch tensor: tokens torch.Size([4, 32768]) +batch tensor: labels torch.Size([4, 32768]) +batch tensor: loss_mask torch.Size([4, 32768]) +batch tensor: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor: position_ids torch.Size([4, 32768]) +batch tensor after cp: tokens torch.Size([4, 32768]) +batch tensor after cp: labels torch.Size([4, 32768]) +batch tensor after cp: loss_mask torch.Size([4, 32768]) +batch tensor after cp: attention_mask torch.Size([4, 1, 32768, 32768]) +batch tensor after cp: position_ids torch.Size([4, 32768]) +Start exporting trace 11 +Done exporting trace 11 +(min, max) time across ranks (ms): + evaluate .......................................: (638.89, 639.09) +---------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on test set | lm loss value: 1.358201E+01 | lm loss PPL: 7.917604E+05 | +---------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Running ctx_length=12288, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 12288 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 12288 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 12288 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 12288 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.046 seconds +> compiling and loading fused kernels ... +>>> done with compiling and loading fused kernels. Compilation time: 2.582 seconds +time to initialize megatron (seconds): 7.815 +[after megatron is initialized] datetime: 2025-06-21 21:24:01 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 120649216 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding>>> embedding + +>>> decoder +>>> decoder>>> output_layer + +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 120649216 + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 120649216 + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 120649216 > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 120649216 + +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 120649216 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 120649216 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 120649216 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (120649216 elements, 120649216 padded size): + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.final_layernorm.bias + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.embedding.position_embeddings.weight + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.final_layernorm.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.weight +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (3.22, 3.34) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:24:02 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=12288, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.004512 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 5549 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001827 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 5546 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001678 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 5557 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:24:02 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (672.74, 681.00) + train/valid/test-data-iterators-setup ..........: (20.55, 112.00) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:24:02 +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:24:10] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 8582.3 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 2] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 45584.0 | max reserved: 45584.0 +[Rank 3] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44048.0 | max reserved: 44048.0[Rank 4] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44624.0 | max reserved: 44624.0 + +[Rank 1] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44048.0 | max reserved: 44048.0[Rank 6] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44048.0 | max reserved: 44048.0 + +[Rank 7] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44432.0 | max reserved: 44432.0 +[Rank 5] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44048.0 | max reserved: 44048.0 +[Rank 0] (after 1 iterations) memory (MB) | allocated: 10740.85595703125 | max allocated: 40669.55419921875 | reserved: 44624.0 | max reserved: 44624.0 +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:24:12] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 2113.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:24:14] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 2071.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:24:17] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 2069.1 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:24:19] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 2071.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:24:21] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 2089.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:24:23] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 2070.4 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:24:25] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 2120.2 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:24:27] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 2074.4 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +batch tensor: tokens torch.Size([4, 49152]) +batch tensor: labels torch.Size([4, 49152]) +batch tensor: loss_mask torch.Size([4, 49152]) +batch tensor: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor: position_ids torch.Size([4, 49152]) +batch tensor after cp: tokens torch.Size([4, 49152]) +batch tensor after cp: labels torch.Size([4, 49152]) +batch tensor after cp: loss_mask torch.Size([4, 49152]) +batch tensor after cp: attention_mask torch.Size([4, 1, 49152, 49152]) +batch tensor after cp: position_ids torch.Size([4, 49152]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:24:29] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 2092.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:24:29 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.03065180778503418 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.030994415283203125 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.031183719635009766 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.03133106231689453 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.03136754035949707 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.03230786323547363 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.033994436264038086 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.03727245330810547 to prepare state dict for ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +Running ctx_length=16384, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 16384 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 16384 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 16384 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 16384 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +INFO:megatron.training.initialize:Setting logging level to 0 +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +INFO:megatron.training.initialize:Setting logging level to 0 +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.046 seconds +> compiling and loading fused kernels ... +>>> done with compiling and loading fused kernels. Compilation time: 2.638 seconds +time to initialize megatron (seconds): 7.043 +[after megatron is initialized] datetime: 2025-06-21 21:25:46 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 137426432 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 137426432 + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 137426432 + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 137426432 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 137426432 + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 137426432 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 137426432 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 137426432 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (137426432 elements, 137426432 padded size): + module.decoder.final_layernorm.bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.final_layernorm.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.embedding.word_embeddings.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.embedding.position_embeddings.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (3.58, 3.68) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:25:47 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=16384, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.005091 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 4162 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001775 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 4160 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001480 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 4167 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:25:47 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (813.07, 827.01) + train/valid/test-data-iterators-setup ..........: (37.44, 125.01) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:25:47 +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:25:59] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 12014.1 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 2] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 62580.0 | max reserved: 62580.0 +[Rank 6] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 64628.0 | max reserved: 64628.0[Rank 3] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 64628.0 | max reserved: 64628.0 + +[Rank 5] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 62580.0 | max reserved: 62580.0 +[Rank 1] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 64628.0 | max reserved: 64628.0 +[Rank 0] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 62580.0 | max reserved: 62580.0 +[Rank 7] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 62580.0 | max reserved: 62580.0 +[Rank 4] (after 1 iterations) memory (MB) | allocated: 18101.72607421875 | max allocated: 58116.98681640625 | reserved: 62580.0 | max reserved: 62580.0 +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:26:02] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 3412.5 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:26:06] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 3454.9 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:26:09] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 3417.3 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:26:13] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 3441.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:26:16] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 3425.7 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:26:20] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 3480.4 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:26:23] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 3489.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:26:27] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 3478.1 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:26:30] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 3430.1 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:26:30 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.031345367431640625 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.03149843215942383 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.03169083595275879 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.031677961349487305 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.03170275688171387 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.031714677810668945 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.03174185752868652 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.03576397895812988 to prepare state dict for ckpt +WARNING:megatron.core.dist_checkpointing.serialization:Overwriting old incomplete / corrupted checkpoint... +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.3674821853637695 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.3674907684326172 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.3675310611724854 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.3675425052642822 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.367002010345459 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.367412805557251 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 1.367767095565796 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 0.007026195526123047 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, plan time: 0.005448818206787109 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, plan time: 0.005511283874511719 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, plan time: 0.005213260650634766 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, plan time: 0.0051915645599365234 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, plan time: 0.005080223083496094 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, plan time: 0.005156040191650391 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, plan time: 0.00515437126159668 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.7769341 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.7769427 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.776944 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.7769413 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.7769504 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.776949 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.776957 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 8.678436279296875e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.131431579589844e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.202957153320312e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.703636169433594e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.369850158691406e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.274482727050781e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010538101196289062 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, plan time: 0.007027864456176758 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541192.7816997 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.0001049041748046875 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04643082618713379 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04643058776855469 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04650521278381348 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.823818 rank: 3, write(async) time: 0.046872854232788086 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8238177 rank: 1, write(async) time: 0.046872615814208984 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8238914 rank: 6, write(async) time: 0.04695558547973633 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.0483705997467041 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8257344 rank: 2, write(async) time: 0.048790693283081055 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.049704551696777344 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8271186 rank: 7, write(async) time: 0.05016946792602539 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.0517885684967041 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8291755 rank: 4, write(async) time: 0.05221962928771973 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.05315732955932617 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8306296 rank: 5, write(async) time: 0.05367755889892578 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.053144216537475586 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541192.8353603 rank: 0, write(async) time: 0.05365705490112305 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 1.6927719116210938e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 1.4781951904296875e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 1.6689300537109375e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 1.4066696166992188e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 1.71661376953125e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 1.4543533325195312e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 1.621246337890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 0.024404048919677734 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 0.025645732879638672 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 0.025215864181518555 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 0.026580333709716797 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 0.02644062042236328 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 0.03064894676208496 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 0.03379416465759277 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 1.6689300537109375e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 0.029697656631469727 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214552576, before: 1615122432, after: 1829675008 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214642688, before: 1625387008, after: 1840029696 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214700032, before: 1616838656, after: 1831538688 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214577152, before: 1625206784, after: 1839783936 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216641536, before: 1625387008, after: 1842028544 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214573056, before: 1633984512, after: 1848557568 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216727552, before: 1633984512, after: 1850712064 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 215900160, before: 1625206784, after: 1841106944 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216559616, before: 1615122432, after: 1831682048 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216707072, before: 1616838656, after: 1833545728 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216682496, before: 1613819904, after: 1830502400 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.9308517, rank: 2, write(sync,parallel): 0.890728235244751 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214478848, before: 1609752576, after: 1824231424 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.9449039, rank: 7, write(sync,parallel): 0.9008100032806396 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216444928, before: 1609789440, after: 1826234368 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.950929, rank: 1, write(sync,parallel): 0.9179232120513916 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.9521396, rank: 3, write(sync,parallel): 0.9120340347290039 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214700032, before: 1613819904, after: 1828519936 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.9620035, rank: 6, write(sync,parallel): 0.921734094619751 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.95s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.96s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.98s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.98s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.9924722, rank: 5, write(sync,parallel): 0.9382729530334473 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.99s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541193.9999688, rank: 4, write(sync,parallel): 0.9515547752380371 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 1.01s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 1.03s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 417918976, before: 1882648576, after: 2300567552 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 416251904, before: 1882648576, after: 2298900480 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541194.5838678, rank: 0, write(sync,parallel): 1.3033010959625244 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 1.37s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6201031, 2, gather: 0.6561181545257568 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6201074, 1, gather: 0.6348819732666016 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6201725, 3, gather: 0.6322641372680664 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6203802, 4, gather: 0.5802056789398193 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6204178, 6, gather: 0.6214685440063477 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6204462, 7, gather: 0.641338586807251 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.620844, 5, gather: 0.5905380249023438 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6229954, 0, gather: 0.005300045013427734 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541194.6343362, metadata_write: 0.011212348937988281 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.6724s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.6510s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.6062s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.6373s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0191s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.6573s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.5963s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.6488s + successfully saved checkpoint from iteration 10 to gpt-checkpoint [ t 1/8, p 1/1 ] +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.0019981861114501953 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.0019714832305908203 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.001990795135498047 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.0019626617431640625 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.00199127197265625 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.0019919872283935547 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.0019829273223876953 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.0019609928131103516 to finalize ckpt save +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 10 +Done exporting trace 10 +(min, max) time across ranks (ms): + evaluate .......................................: (4121.85, 4122.13) +---------------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on validation set | lm loss value: 1.228851E+01 | lm loss PPL: 2.171866E+05 | +---------------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +batch tensor: tokens torch.Size([4, 65536]) +batch tensor: labels torch.Size([4, 65536]) +batch tensor: loss_mask torch.Size([4, 65536]) +batch tensor: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor: position_ids torch.Size([4, 65536]) +batch tensor after cp: tokens torch.Size([4, 65536]) +batch tensor after cp: labels torch.Size([4, 65536]) +batch tensor after cp: loss_mask torch.Size([4, 65536]) +batch tensor after cp: attention_mask torch.Size([4, 1, 65536, 65536]) +batch tensor after cp: position_ids torch.Size([4, 65536]) +Start exporting trace 11 +Done exporting trace 11 +(min, max) time across ranks (ms): + evaluate .......................................: (2508.12, 2508.32) +---------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on test set | lm loss value: 1.228851E+01 | lm loss PPL: 2.171866E+05 | +---------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Running ctx_length=24576, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 24576 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 24576 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 24576 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 24576 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.044 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +>>> done with compiling and loading fused kernels. Compilation time: 2.479 seconds +time to initialize megatron (seconds): 7.086 +[after megatron is initialized] datetime: 2025-06-21 21:27:19 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 170980864 + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 170980864 + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 170980864 +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 170980864 + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 170980864 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 170980864 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 170980864 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 170980864 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (170980864 elements, 170980864 padded size): + module.decoder.final_layernorm.bias + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.embedding.position_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.final_layernorm.weight + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.layers.0.mlp.linear_fc1.bias +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (3.10, 3.12) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:27:20 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=24576, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.005042 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 2774 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001786 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 2773 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001492 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 2778 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:27:20 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (1192.44, 1212.58) + train/valid/test-data-iterators-setup ..........: (19.01, 135.11) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:27:20 +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 0 +Done exporting trace 0 + [2025-06-21 21:27:35] iteration 1/ 10 | consumed samples: 1 | elapsed time per iteration (ms): 15013.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 4294967296.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 0.35 +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 0.56 +Number of parameters in most loaded shard in billions: 0.0703 +Theoretical memory footprints: weight and optimizer=1206.09 MB +[Rank 1] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0[Rank 5] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0[Rank 4] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0 + + +[Rank 6] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0[Rank 7] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0 + +[Rank 2] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0 +[Rank 0] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0 +[Rank 3] (after 1 iterations) memory (MB) | allocated: 38970.22607421875 | max allocated: 99166.61669921875 | reserved: 105836.0 | max reserved: 105836.0 +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 1 +Done exporting trace 1 + [2025-06-21 21:27:43] iteration 2/ 10 | consumed samples: 2 | elapsed time per iteration (ms): 7719.7 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 2147483648.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 2 +Done exporting trace 2 + [2025-06-21 21:27:50] iteration 3/ 10 | consumed samples: 3 | elapsed time per iteration (ms): 7690.9 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 1073741824.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 3 +Done exporting trace 3 + [2025-06-21 21:27:58] iteration 4/ 10 | consumed samples: 4 | elapsed time per iteration (ms): 7538.8 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 536870912.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 4 +Done exporting trace 4 + [2025-06-21 21:28:05] iteration 5/ 10 | consumed samples: 5 | elapsed time per iteration (ms): 7469.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 268435456.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304])batch tensor after cp: + tokensbatch tensor after cp: attention_masktorch.Size([4, 98304]) +torch.Size([4, 1, 98304, 98304])batch tensor after cp: + labelsbatch tensor after cp: position_idstorch.Size([4, 98304]) +torch.Size([4, 98304])batch tensor after cp: + loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 5 +Done exporting trace 5 + [2025-06-21 21:28:13] iteration 6/ 10 | consumed samples: 6 | elapsed time per iteration (ms): 7476.7 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 134217728.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 6 +Done exporting trace 6 + [2025-06-21 21:28:21] iteration 7/ 10 | consumed samples: 7 | elapsed time per iteration (ms): 7554.4 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 67108864.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 7 +Done exporting trace 7 + [2025-06-21 21:28:28] iteration 8/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 7413.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 33554432.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 8 +Done exporting trace 8 + [2025-06-21 21:28:35] iteration 9/ 10 | consumed samples: 9 | elapsed time per iteration (ms): 7457.6 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 16777216.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 9 +Done exporting trace 9 + [2025-06-21 21:28:43] iteration 10/ 10 | consumed samples: 10 | elapsed time per iteration (ms): 7508.0 | learning rate: 0.000000E+00 | global batch size: 1 | loss scale: 8388608.0 | number of skipped iterations: 1 | number of nan iterations: 0 | +[after training is done] datetime: 2025-06-21 21:28:43 +saving checkpoint at iteration 10 to gpt-checkpoint in torch_dist format +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.03838968276977539 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.03858137130737305 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.03861188888549805 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.038706064224243164 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.038719892501831055 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.03972482681274414 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.039731740951538086 to prepare state dict for ckpt +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.04030036926269531 to prepare state dict for ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:Apply save parallelization +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.177408456802368 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.177398920059204 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.1774814128875732 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.1774985790252686 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.1775217056274414 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.177586317062378 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 2.1775686740875244 +DEBUG:megatron.core.dist_checkpointing.strategies.fully_parallel:parallel save sharding, time: 0.0071256160736083984 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, starting state dict save +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:verifying reuse of global metadata +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:loaded global metadata reuse verification: no loaded plans passed +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 6, plan time: 0.006767988204956055 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 7, plan time: 0.0050201416015625 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 1, plan time: 0.006754398345947266 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 4, plan time: 0.006791591644287109 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 5, plan time: 0.005375862121582031 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 2, plan time: 0.0061299800872802734 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5376685 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5376644 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5376773 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5376773 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5376792 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5376828 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 3, plan time: 0.005210161209106445 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5377214 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.894371032714844e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.942054748535156e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.083747863769531e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.226799011230469e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010180473327636719 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 9.72747802734375e-05 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00010251998901367188 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:rank: 0, plan time: 0.007717609405517578 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:thread_count: 2, time: 1750541326.5427105 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:bucket_prep, time: 0.00011157989501953125 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04523921012878418 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.5833642 rank: 4, write(async) time: 0.0456850528717041 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04611921310424805 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.5842462 rank: 7, write(async) time: 0.046575307846069336 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04715085029602051 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.5853302 rank: 6, write(async) time: 0.04766249656677246 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.047528743743896484 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.5856915 rank: 3, write(async) time: 0.047968149185180664 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04813194274902344 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.0481715202331543 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.5862932 rank: 5, write(async) time: 0.04861140251159668 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.5863376 rank: 2, write(async) time: 0.048636436462402344 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.04832935333251953 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.586473 rank: 1, write(async) time: 0.04879403114318848 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:D2H and push, time: 0.057624101638793945 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541326.6008697 rank: 0, write(async) time: 0.05815601348876953 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 1.621246337890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 1.7404556274414062e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 1.621246337890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 1.71661376953125e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 1.71661376953125e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 1.621246337890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 1.6450881958007812e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, takes 0.027580976486206055 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, takes 0.025325536727905273 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, takes 0.028856277465820312 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, takes 0.026805639266967773 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, takes 0.027535676956176758 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, takes 0.026509761810302734 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, takes 0.029981136322021484 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 1.3828277587890625e-05 to finish D2H +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, takes 0.032510995864868164 to schedule async ckpt +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 1, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 2, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 3, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 4, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 7, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 5, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 6, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:rank: 0, joining self.process +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collecting worker results... +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 started +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214581248, before: 1611767808, after: 1826349056 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214634496, before: 1625776128, after: 1840410624 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216526848, before: 1611767808, after: 1828294656 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214634496, before: 1649946624, after: 1864581120 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214585344, before: 1611882496, after: 1826467840 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214462464, before: 1633738752, after: 1848201216 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216637440, before: 1625776128, after: 1842413568 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214433792, before: 1618092032, after: 1832525824 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216670208, before: 1649942528, after: 1866612736 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 215855104, before: 1611882496, after: 1827737600 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 214413312, before: 1612304384, after: 1826717696 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.3622339, rank: 7, write(sync,parallel): 0.5683159828186035 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216477696, before: 1633738752, after: 1850216448 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216616960, before: 1612304384, after: 1828921344 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 216535040, before: 1618092032, after: 1834627072 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.3907478, rank: 6, write(sync,parallel): 0.5960125923156738 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.63s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.4004905, rank: 2, write(sync,parallel): 0.5966086387634277 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.4048283, rank: 4, write(sync,parallel): 0.6077392101287842 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.4138594, rank: 3, write(sync,parallel): 0.6074683666229248 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.4159393, rank: 1, write(sync,parallel): 0.6064801216125488 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541327.4200125, rank: 5, write(sync,parallel): 0.6074962615966797 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.67s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.67s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 0.68s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1 consumed: 518569984, before: 1884504064, after: 2403074048 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:0 consumed: 516956160, before: 1884487680, after: 2401443840 +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:FileSystemWriterAsync: collected worker results successfully +DEBUG:megatron.core.dist_checkpointing.strategies.filesystem_async:1750541328.5343385, rank: 0, write(sync,parallel): 1.3298144340515137 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:TemporalAsyncCaller: Async process join finished after 1.42s from forking +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5868359, 1, gather: 1.1309428215026855 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5869915, 4, gather: 1.1387341022491455 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.587006, 2, gather: 1.1463701725006104 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5871289, 5, gather: 1.124898910522461 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5872228, 3, gather: 1.1331896781921387 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5873969, 7, gather: 1.1910743713378906 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5876358, 6, gather: 1.1536352634429932 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.5911682, 0, gather: 0.006685972213745117 +DEBUG:megatron.core.dist_checkpointing.strategies.state_dict_saver:1750541328.6034317, metadata_write: 0.012113094329833984 +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 0.0220s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.1575s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.1497s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.1518s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.1651s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.1435s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.2093s +DEBUG:megatron.core.dist_checkpointing.strategies.async_utils:finalize took 1.1723s + successfully saved checkpoint from iteration 10 to gpt-checkpoint [ t 1/8, p 1/1 ] +DEBUG:megatron.training.checkpointing:rank: 3, takes 0.0025305747985839844 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 1, takes 0.0025467872619628906 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 6, takes 0.002429962158203125 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 7, takes 0.002528667449951172 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 2, takes 0.00254058837890625 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 4, takes 0.0025615692138671875 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 5, takes 0.0025315284729003906 to finalize ckpt save +DEBUG:megatron.training.checkpointing:rank: 0, takes 0.0025649070739746094 to finalize ckpt save +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 10 +Done exporting trace 10 +(min, max) time across ranks (ms): + evaluate .......................................: (6891.36, 6891.73) +---------------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on validation set | lm loss value: 1.268938E+01 | lm loss PPL: 3.242846E+05 | +---------------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Evaluating on 1 samples +Evaluating iter 1/1 +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +batch tensor: tokens torch.Size([4, 98304]) +batch tensor: labels torch.Size([4, 98304]) +batch tensor: loss_mask torch.Size([4, 98304]) +batch tensor: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor: position_ids torch.Size([4, 98304]) +batch tensor after cp: tokens torch.Size([4, 98304]) +batch tensor after cp: labels torch.Size([4, 98304]) +batch tensor after cp: loss_mask torch.Size([4, 98304]) +batch tensor after cp: attention_mask torch.Size([4, 1, 98304, 98304]) +batch tensor after cp: position_ids torch.Size([4, 98304]) +Start exporting trace 11 +Done exporting trace 11 +(min, max) time across ranks (ms): + evaluate .......................................: (5351.53, 5352.41) +---------------------------------------------------------------------------------------------------------- + validation loss at iteration 10 on test set | lm loss value: 1.268938E+01 | lm loss PPL: 3.242846E+05 | +---------------------------------------------------------------------------------------------------------- +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +WARNING:megatron.core.rerun_state_machine:Setting RerunStateMachine mode RerunMode.DISABLED +Running ctx_length=32768, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 32768 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 32768 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 32768 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 32768 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.045 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +>>> done with compiling and loading fused kernels. Compilation time: 2.920 seconds +time to initialize megatron (seconds): 7.817 +[after megatron is initialized] datetime: 2025-06-21 21:29:39 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 204535296 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 204535296 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 204535296 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 204535296 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 204535296 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 204535296 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 204535296 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (204535296 elements, 204535296 padded size): + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.embedding.word_embeddings.weight + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.final_layernorm.bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.final_layernorm.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.embedding.position_embeddings.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.layers.1.mlp.linear_fc1.bias +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 204535296 +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (2.99, 3.15) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:29:41 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=32768, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.005157 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 2081 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001757 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 2080 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001445 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 2083 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:29:41 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (1569.98, 1571.72) + train/valid/test-data-iterators-setup ..........: (17.16, 111.85) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:29:41 +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +batch tensor: tokens torch.Size([4, 131072]) +batch tensor: labels torch.Size([4, 131072]) +batch tensor: loss_mask torch.Size([4, 131072]) +batch tensor: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor: position_ids torch.Size([4, 131072]) +batch tensor after cp: tokens torch.Size([4, 131072]) +batch tensor after cp: labels torch.Size([4, 131072]) +batch tensor after cp: loss_mask torch.Size([4, 131072]) +batch tensor after cp: attention_mask torch.Size([4, 1, 131072, 131072]) +batch tensor after cp: position_ids torch.Size([4, 131072]) +Running ctx_length=40960, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 40960 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3 +INFO:megatron.training.initialize:Setting logging level to 0 +using world size: 8, data-parallel size: 1, context-parallel size: 1, hierarchical context-parallel sizes: Nonetensor-model-parallel size: 8, encoder-tensor-model-parallel size: 0, pipeline-model-parallel size: 1, encoder-pipeline-model-parallel size: 0 +Number of virtual stages per pipeline stage: None +WARNING: Setting args.check_for_nan_in_loss_and_grad to False since dynamic loss scaling is being used +using torch.float16 for parameters ... +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. False + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. True + add_position_embedding .......................... True + add_qkv_bias .................................... True + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... False + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.1 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ False + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ True + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 1 + data_path ....................................... None + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + deprecated_use_mcore_models ..................... False + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 2 + encoder_pipeline_model_parallel_size ............ 0 + encoder_seq_length .............................. 40960 + encoder_tensor_model_parallel_size .............. 0 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 16 + eval_iters ...................................... 1 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 1 + expert_tensor_parallel_size ..................... 8 + external_cuda_graph ............................. False + ffn_hidden_size ................................. 16384 + finetune ........................................ False + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ True + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. None + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... delayed + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 1 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.1 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.02 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kv_channels ..................................... 64 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ gpt-checkpoint + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. False + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... 0 + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0005 + lr_decay_iters .................................. 150000 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 2 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... True + max_position_embeddings ......................... 40960 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 0.0 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... True + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.0 + moe_enable_deepep ............................... False + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. None + moe_grouped_gemm ................................ False + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ None + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 2 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... allgather + moe_token_drop_policy ........................... probs + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 1.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... LayerNorm + num_attention_heads ............................. 64 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... None + num_layers ...................................... 2 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 16 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ False + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.float16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_size .................... 1 + pipeline_model_parallel_split_rank .............. None + position_embedding_type ......................... learned_absolute + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... disabled + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + rope_scaling_factor ............................. 8.0 + rotary_base ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ gpt-checkpoint + save_interval ................................... 16 + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 40960 + sequence_parallel ............................... False + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... None + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... False + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. False + tensor_model_parallel_size ...................... 8 + tensorboard_dir ................................. tensorboard-logs/ + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 10 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. False + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... False + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1 +> building GPT2BPETokenizer tokenizer ... +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 + > padded vocab (size: 50257) with 943 dummy tokens (new size: 51200) +INFO:megatron.training.initialize:Setting logging level to 0 +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.DISABLED +> initializing torch distributed ... +INFO:megatron.training.initialize:Setting logging level to 0 +> initialized tensor model parallel with size 8 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +> compiling dataset index builder ... +make: Entering directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +WARNING: TensorBoard writing requested but is not available (are you using PyTorch 1.1.0 or later?), no TensorBoard logs will be written. +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +INFO:megatron.training.initialize:Setting logging level to 0 +make: Nothing to be done for 'default'. +make: Leaving directory '/mnt/weka/home/hao.zhang/junda/attnserver-megatron/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.058 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +>>> done with compiling and loading fused kernels. Compilation time: 2.482 seconds +time to initialize megatron (seconds): 7.134 +[after megatron is initialized] datetime: 2025-06-21 21:30:35 +building GPT model ... +>>> embedding +>>> decoder +>>> output_layer +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 238089728 + > number of parameters on (tensor, pipeline) model parallel rank (4, 0): 238089728 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (3, 0): 238089728 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (1, 0): 238089728 +INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=False, overlap_grad_reduce=False, overlap_param_gather=False, align_param_gather=False, use_distributed_optimizer=False, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=None, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 1 +Params for bucket 1 (238089728 elements, 238089728 padded size): + module.decoder.layers.1.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.0.mlp.linear_fc2.bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.1.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.final_layernorm.bias + module.decoder.layers.1.mlp.linear_fc2.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.embedding.word_embeddings.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_bias + module.decoder.layers.0.self_attention.linear_proj.bias + module.decoder.layers.1.mlp.linear_fc1.bias + module.decoder.final_layernorm.weight + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.bias + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.1.mlp.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_proj.bias + module.decoder.layers.1.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.mlp.linear_fc1.layer_norm_bias + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.embedding.position_embeddings.weight +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0005, min_lr=0.0, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp16=True, bf16=False, params_dtype=torch.float16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=False, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (6, 0): 238089728 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (5, 0): 238089728 +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (7, 0): 238089728 +>>> embedding +>>> decoder +>>> output_layer + > number of parameters on (tensor, pipeline) model parallel rank (2, 0): 238089728 +WARNING: could not find the metadata file gpt-checkpoint/latest_checkpointed_iteration.txt + will not load any checkpoints and will start from random +(min, max) time across ranks (ms): + load-checkpoint ................................: (4.86, 5.11) +[after model, optimizer, and learning rate scheduler are built] datetime: 2025-06-21 21:30:36 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 10 + validation: 1 + test: 1 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let mock = True, as both blend and blend_per_split are None +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split = 1,1,1, an arbitrarily even split, as mock is True +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building MockGPTDataset splits with sizes=(10, 1, 1) and config=GPTDatasetConfig(random_seed=1234, sequence_length=40960, blend=None, blend_per_split=None, split='1,1,1', split_matrix=[(0, 0.3333333333333333), (0.3333333333333333, 0.6666666666666666), (0.6666666666666666, 1.0)], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=True, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset train indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.004652 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 1664 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset valid indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001627 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 1664 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +INFO:megatron.core.datasets.gpt_dataset:Build and save the MockGPTDataset test indices +DEBUG:megatron.core.datasets.gpt_dataset:> separate_final_epoch: False +WARNING:megatron.core.datasets.gpt_dataset:Unable to save MockGPTDataset indexes because path_to_cache is None +DEBUG:megatron.core.datasets.gpt_dataset: > time elapsed: 0.001342 seconds +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 1667 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2025-06-21 21:30:37 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (1920.40, 1939.87) + train/valid/test-data-iterators-setup ..........: (21.91, 117.38) +training ... +Setting rerun_state_machine.current_iteration to 0... +[before the start of training step] datetime: 2025-06-21 21:30:37 +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +batch tensor: tokens torch.Size([4, 163840]) +batch tensor: labels torch.Size([4, 163840]) +batch tensor: loss_mask torch.Size([4, 163840]) +batch tensor: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor: position_ids torch.Size([4, 163840]) +batch tensor after cp: tokens torch.Size([4, 163840]) +batch tensor after cp: labels torch.Size([4, 163840]) +batch tensor after cp: loss_mask torch.Size([4, 163840]) +batch tensor after cp: attention_mask torch.Size([4, 1, 163840, 163840]) +batch tensor after cp: position_ids torch.Size([4, 163840]) +Running ctx_length=49152, TP_SIZE=8, CP_SIZE=1, BATCH_SIZE=4 +Cleaning up checkpoint directory: gpt-checkpoint +-------------------------------- +CTX_LENGTH: 49152 +TP_SIZE: 8 +CP_SIZE: 1 +CHECKPOINT_PATH: gpt-checkpoint +PWD: /mnt/weka/home/hao.zhang/junda/attnserver-megatron +-------------------------------- +/mnt/weka/home/hao.zhang/conda/miniconda/envs/junda-attnserver/bin/python3