diff --git "a/slurm.out" "b/slurm.out" new file mode 100644--- /dev/null +++ "b/slurm.out" @@ -0,0 +1,568 @@ +3: W0902 18:58:55.388000 284480 torch/distributed/run.py:792] +3: W0902 18:58:55.388000 284480 torch/distributed/run.py:792] ***************************************** +3: W0902 18:58:55.388000 284480 torch/distributed/run.py:792] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +3: W0902 18:58:55.388000 284480 torch/distributed/run.py:792] ***************************************** +0: W0902 18:58:55.639000 2134898 torch/distributed/run.py:792] +0: W0902 18:58:55.639000 2134898 torch/distributed/run.py:792] ***************************************** +0: W0902 18:58:55.639000 2134898 torch/distributed/run.py:792] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +0: W0902 18:58:55.639000 2134898 torch/distributed/run.py:792] ***************************************** +1: W0902 18:58:55.819000 3856158 torch/distributed/run.py:792] +1: W0902 18:58:55.819000 3856158 torch/distributed/run.py:792] ***************************************** +1: W0902 18:58:55.819000 3856158 torch/distributed/run.py:792] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +1: W0902 18:58:55.819000 3856158 torch/distributed/run.py:792] ***************************************** +2: W0902 18:58:56.452000 3393108 torch/distributed/run.py:792] +2: W0902 18:58:56.452000 3393108 torch/distributed/run.py:792] ***************************************** +2: W0902 18:58:56.452000 3393108 torch/distributed/run.py:792] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +2: W0902 18:58:56.452000 3393108 torch/distributed/run.py:792] ***************************************** +3: [2025-09-02 18:59:20,646] [INFO] [axolotl.utils.schemas.validation.check_eval_packing:119] [PID:284556] [RANK:0] explicitly setting `eval_sample_packing` to match `sample_packing` +3: [2025-09-02 18:59:20,646] [INFO] [axolotl.utils.schemas.validation.hint_sample_packing_padding:218] [PID:284556] [RANK:0] Setting `pad_to_sequence_len: true` to prevent memory leaks when sample_packing +1: [2025-09-02 18:59:21,727] [INFO] [axolotl.utils.schemas.validation.check_eval_packing:119] [PID:3856235] [RANK:0] explicitly setting `eval_sample_packing` to match `sample_packing` +1: [2025-09-02 18:59:21,728] [INFO] [axolotl.utils.schemas.validation.hint_sample_packing_padding:218] [PID:3856235] [RANK:0] Setting `pad_to_sequence_len: true` to prevent memory leaks when sample_packing +0: [2025-09-02 18:59:23,380] [INFO] [axolotl.utils.schemas.validation.check_eval_packing:119] [PID:2134977] [RANK:0] explicitly setting `eval_sample_packing` to match `sample_packing` +0: [2025-09-02 18:59:23,380] [INFO] [axolotl.utils.schemas.validation.hint_sample_packing_padding:218] [PID:2134977] [RANK:0] Setting `pad_to_sequence_len: true` to prevent memory leaks when sample_packing +2: [2025-09-02 18:59:23,899] [INFO] [axolotl.utils.schemas.validation.check_eval_packing:119] [PID:3393184] [RANK:0] explicitly setting `eval_sample_packing` to match `sample_packing` +2: [2025-09-02 18:59:23,899] [INFO] [axolotl.utils.schemas.validation.hint_sample_packing_padding:218] [PID:3393184] [RANK:0] Setting `pad_to_sequence_len: true` to prevent memory leaks when sample_packing +3: [2025-09-02 18:59:25,238] [INFO] [axolotl.utils.data.sft._load_raw_datasets:314] [PID:284557] [RANK:1] Loading raw datasets... +3: [2025-09-02 18:59:25,510] [INFO] [axolotl.utils.data.wrappers.get_dataset_wrapper:88] [PID:284557] [RANK:1] Loading dataset: /lustre/fswork/projects/rech/qwv/udv55np/dataset/math/hf/no_thinking_text/generator/default-d32b2cae8ea7e541/0.0.0 with base_type: chat_template and prompt_style: None +0: [2025-09-02 18:59:27,157] [INFO] [axolotl.cli.config.load_cfg:245] [PID:2134977] [RANK:0] config: +0: { +0: "activation_offloading": false, +0: "auto_resume_from_checkpoints": true, +0: "axolotl_config_path": "/lustre/fswork/projects/rech/dgo/udv55np/train/tmp/1756826506262014665.yaml", +0: "base_model": "/lustre/fswork/projects/rech/qwv/udv55np/Qwen/Qwen2.5-14B_ift", +0: "base_model_config": "/lustre/fswork/projects/rech/qwv/udv55np/Qwen/Qwen2.5-14B_ift", +0: "batch_size": 16, +0: "bf16": true, +0: "capabilities": { +0: "bf16": true, +0: "compute_capability": "sm_90", +0: "fp8": false, +0: "n_gpu": 16, +0: "n_node": 1 +0: }, +0: "chat_template": "qwen_25", +0: "context_parallel_size": 1, +0: "dataloader_num_workers": 16, +0: "dataloader_pin_memory": true, +0: "dataloader_prefetch_factor": 256, +0: "dataset_prepared_path": "/lustre/fsn1/projects/rech/dgo/udv55np/dataset_math/Qwen3-235B-A22B/0", +0: "dataset_processes": 192, +0: "datasets": [ +0: { +0: "chat_template": "tokenizer_default", +0: "field_messages": "conversations", +0: "message_property_mappings": { +0: "content": "content", +0: "role": "role" +0: }, +0: "path": "/lustre/fswork/projects/rech/qwv/udv55np/dataset/math/hf/no_thinking_text/generator/default-d32b2cae8ea7e541/0.0.0", +0: "trust_remote_code": false, +0: "type": "chat_template" +0: } +0: ], +0: "ddp": true, +0: "deepspeed": { +0: "bf16": { +0: "enabled": true +0: }, +0: "gradient_accumulation_steps": "auto", +0: "gradient_clipping": "auto", +0: "train_batch_size": "auto", +0: "train_micro_batch_size_per_gpu": "auto", +0: "wall_clock_breakdown": false, +0: "zero_optimization": { +0: "contiguous_gradients": true, +0: "overlap_comm": true, +0: "reduce_bucket_size": "auto", +0: "stage": 3, +0: "stage3_gather_16bit_weights_on_model_save": true, +0: "stage3_param_persistence_threshold": "auto", +0: "stage3_prefetch_bucket_size": "auto", +0: "sub_group_size": 0 +0: } +0: }, +0: "device": "cuda:0", +0: "device_map": { +0: "": 0 +0: }, +0: "dion_rank_fraction": 1.0, +0: "dion_rank_multiple_of": 1, +0: "env_capabilities": { +0: "torch_version": "2.6.0" +0: }, +0: "eval_batch_size": 1, +0: "eval_causal_lm_metrics": [ +0: "sacrebleu", +0: "comet", +0: "ter", +0: "chrf" +0: ], +0: "eval_max_new_tokens": 128, +0: "eval_sample_packing": true, +0: "eval_table_size": 0, +0: "evals_per_epoch": 0, +0: "flash_attention": true, +0: "fp16": false, +0: "gradient_accumulation_steps": 1, +0: "gradient_checkpointing": true, +0: "gradient_checkpointing_kwargs": { +0: "use_reentrant": true +0: }, +0: "learning_rate": 1e-06, +0: "lisa_layers_attribute": "model.layers", +0: "load_best_model_at_end": false, +0: "load_in_4bit": false, +0: "load_in_8bit": false, +0: "local_rank": 0, +0: "logging_steps": 10, +0: "lora_dropout": 0.0, +0: "loraplus_lr_embedding": 1e-06, +0: "lr_scheduler": "warmup_stable_decay", +0: "lr_scheduler_kwargs": { +0: "min_lr_ratio": 0.1, +0: "num_decay_steps": 300 +0: }, +0: "max_prompt_len": 512, +0: "mean_resizing_embeddings": false, +0: "micro_batch_size": 1, +0: "model_config_type": "qwen2", +0: "num_epochs": 1.0, +0: "optimizer": "adamw_torch_fused", +0: "output_dir": "/lustre/fswork/projects/rech/dgo/udv55np/math/Qwen3-235B-A22B/Qwen2.5-14B_ift/0", +0: "pad_to_sequence_len": true, +0: "pretrain_multipack_attn": true, +0: "pretrain_multipack_buffer_size": 10000, +0: "profiler_steps_start": 0, +0: "qlora_sharded_model_loading": false, +0: "ray_num_workers": 1, +0: "resources_per_worker": { +0: "GPU": 1 +0: }, +0: "sample_packing": true, +0: "sample_packing_bin_size": 200, +0: "sample_packing_group_size": 100000, +0: "save_only_model": false, +0: "save_safetensors": true, +0: "save_steps": 0.2, +0: "save_total_limit": 20, +0: "sequence_len": 16384, +0: "shuffle_before_merging_datasets": false, +0: "shuffle_merged_datasets": true, +0: "skip_prepare_dataset": false, +0: "special_tokens": { +0: "bos_token": "<|im_start|>", +0: "eos_token": "<|im_end|>", +0: "pad_token": "<|endoftext|>" +0: }, +0: "strict": false, +0: "tensor_parallel_size": 1, +0: "tf32": false, +0: "tiled_mlp_use_original_mlp": true, +0: "tokenizer_config": "/lustre/fswork/projects/rech/qwv/udv55np/Qwen/Qwen2.5-14B_ift", +0: "torch_dtype": "torch.bfloat16", +0: "train_on_inputs": false, +0: "trl": { +0: "log_completions": false, +0: "mask_truncated_completions": false, +0: "ref_model_mixup_alpha": 0.9, +0: "ref_model_sync_steps": 64, +0: "scale_rewards": true, +0: "sync_ref_model": false, +0: "use_vllm": false, +0: "vllm_server_host": "0.0.0.0", +0: "vllm_server_port": 8000 +0: }, +0: "use_ray": false, +0: "use_tensorboard": true, +0: "val_set_size": 0.0, +0: "vllm": { +0: "device": "auto", +0: "dtype": "auto", +0: "gpu_memory_utilization": 0.9, +0: "host": "0.0.0.0", +0: "port": 8000 +0: }, +0: "warmup_steps": 150, +0: "weight_decay": 0.0, +0: "world_size": 16 +0: } +0: [2025-09-02 18:59:27,159] [INFO] [axolotl.cli.checks.check_user_token:35] [PID:2134977] [RANK:0] Skipping HuggingFace token verification because HF_HUB_OFFLINE is set to True. Only local files will be used. +3: Tokenizing Prompts (num_proc=192): 0%| | 0/321773 [00:0016384) (num_proc=192): 0%| | 0/321773 [00:0016384) (num_proc=192): 0%| | 1000/321773 [00:01<09:44, 548.90 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 5%|▍ | 15000/321773 [00:01<00:28, 10641.73 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 8%|▊ | 24676/321773 [00:02<00:18, 16454.79 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 10%|▉ | 30732/321773 [00:02<00:15, 18680.29 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 11%|█ | 36112/321773 [00:02<00:13, 21058.71 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 13%|█▎ | 40492/321773 [00:02<00:14, 18996.48 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 14%|█▎ | 43844/321773 [00:03<00:14, 19470.16 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 15%|█▌ | 48548/321773 [00:03<00:11, 23185.99 example +3: s/s] Dropping Long Sequences (>16384) (num_proc=192): 16%|█▌ | 52224/321773 [00:03<00:10, 25109.15 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 54%|█████▍ | 173756/321773 [00:03<00:00, 257188.15 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 66%|██████▋ | 213896/321773 [00:03<00:00, 260962.67 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 78%|███████▊ | 249796/321773 [00:03<00:00, 225600.38 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 87%|████████▋ | 280020/321773 [00:03<00:00, 219928.40 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 96%|█████████▌| 307593/321773 [00:04<00:00, 177091.06 examples/s] Dropping Long Sequences (>16384) (num_proc=192): 100%|██████████| 321773/321773 [00:04<00:00, 68433.68 examples/s] +3: Drop Samples with Zero Trainable Tokens (num_proc=192): 0%| | 0/315947 [00:00