| mcore_gpt: true | |
| micro_batch_size: 1 | |
| global_batch_size: 128 | |
| tensor_model_parallel_size: 8 | |
| pipeline_model_parallel_size: 8 | |
| virtual_pipeline_model_parallel_size: null | |
| encoder_seq_length: 8192 | |
| max_position_embeddings: 8192 | |
| num_layers: 80 | |
| hidden_size: 8192 | |
| ffn_hidden_size: 28672 | |
| num_attention_heads: 64 | |
| init_method_std: 0.02 | |
| use_scaled_init_method: true | |
| hidden_dropout: 0.0 | |
| attention_dropout: 0.0 | |
| ffn_dropout: 0.0 | |
| kv_channels: null | |
| apply_query_key_layer_scaling: true | |
| normalization: rmsnorm | |
| layernorm_epsilon: 1.0e-05 | |
| do_layer_norm_weight_decay: false | |
| make_vocab_size_divisible_by: 128 | |
| pre_process: true | |
| post_process: true | |
| persist_layer_norm: true | |
| bias: false | |
| activation: fast-swiglu | |
| headscale: false | |
| transformer_block_type: pre_ln | |
| openai_gelu: false | |
| normalize_attention_scores: true | |
| position_embedding_type: rope | |
| rotary_percentage: 1.0 | |
| attention_type: multihead | |
| share_embeddings_and_output_weights: false | |
| overlap_p2p_comm: false | |
| batch_p2p_comm: true | |
| num_query_groups: 8 | |
| tokenizer: | |
| library: huggingface | |
| type: meta-llama/Meta-Llama-3-70B | |
| use_fast: true | |
| native_amp_init_scale: 4294967296 | |
| native_amp_growth_interval: 1000 | |
| hysteresis: 2 | |
| fp32_residual_connection: false | |
| fp16_lm_cross_entropy: false | |
| megatron_amp_O2: true | |
| grad_allreduce_chunk_size_mb: 125 | |
| grad_div_ar_fusion: true | |
| gradient_accumulation_fusion: false | |
| bias_activation_fusion: false | |
| bias_dropout_add_fusion: false | |
| masked_softmax_fusion: true | |
| get_attention_mask_from_fusion: true | |
| apply_rope_fusion: false | |
| seed: 1234 | |
| resume_from_checkpoint: null | |
| use_cpu_initialization: false | |
| onnx_safe: false | |
| apex_transformer_log_level: 30 | |
| gradient_as_bucket_view: false | |
| sync_batch_comm: false | |
| activations_checkpoint_granularity: full | |
| activations_checkpoint_method: uniform | |
| activations_checkpoint_num_layers: 1 | |
| num_micro_batches_with_partial_activation_checkpoints: null | |
| activations_checkpoint_layers_per_pipeline: null | |
| sequence_parallel: false | |
| transformer_engine: true | |
| fp8: false | |
| fp8_e4m3: false | |
| fp8_hybrid: true | |
| fp8_margin: 0 | |
| fp8_interval: 1 | |
| fp8_amax_history_len: 1024 | |
| fp8_amax_compute_algo: max | |
| reduce_amax: true | |
| use_emha: false | |
| data: | |
| chat: true | |
| chat_prompt_tokens: | |
| system_turn_start: <extra_id_0> | |
| turn_start: <extra_id_1> | |
| label_start: <extra_id_2> | |
| end_of_turn: ' | |
| ' | |
| end_of_name: ' | |
| ' | |
| sample: true | |
| num_workers: 2 | |
| dataloader_type: single | |
| train_ds: | |
| file_path: /dataset/train.jsonl | |
| global_batch_size: 384 | |
| micro_batch_size: 1 | |
| shuffle: true | |
| memmap_workers: null | |
| max_seq_length: 4096 | |
| min_seq_length: 1 | |
| drop_last: true | |
| concat_sampling_probabilities: null | |
| label_key: output | |
| add_eos: false | |
| add_sep: false | |
| add_bos: false | |
| truncation_field: input | |
| index_mapping_dir: /indexmap_dir | |
| prompt_template: '<extra_id_0>System | |
| {system message} | |
| <extra_id_1>User | |
| {turn 1 user message} | |
| <extra_id_1>Assistant | |
| <extra_id_2>{turn 1 assistant label} | |
| {turn 1 assistant message} | |
| <extra_id_1>User | |
| {turn 2 user message} | |
| <extra_id_1>Assistant | |
| <extra_id_2>{turn 2 assistant label} | |
| {turn 2 assistant message} | |
| <extra_id_1>' | |
| hf_dataset: true | |
| truncation_method: right | |
| validation_ds: | |
| file_path: /dataset/train.jsonl | |
| names: null | |
| global_batch_size: 384 | |
| micro_batch_size: 1 | |
| shuffle: false | |
| memmap_workers: null | |
| max_seq_length: 4096 | |
| min_seq_length: 1 | |
| drop_last: false | |
| label_key: output | |
| add_eos: false | |
| add_sep: false | |
| add_bos: false | |
| write_predictions_to_file: false | |
| output_file_path_prefix: null | |
| truncation_field: input | |
| index_mapping_dir: /indexmap_dir | |
| prompt_template: '<extra_id_0>System | |
| {system message} | |
| <extra_id_1>User | |
| {turn 1 user message} | |
| <extra_id_1>Assistant | |
| <extra_id_2>{turn 1 assistant label} | |
| {turn 1 assistant message} | |
| <extra_id_1>User | |
| {turn 2 user message} | |
| <extra_id_1>Assistant | |
| <extra_id_2>{turn 2 assistant label} | |
| {turn 2 assistant message} | |
| <extra_id_1>' | |
| tokens_to_generate: 32 | |
| hf_dataset: true | |
| truncation_method: right | |
| metric: | |
| name: loss | |
| average: null | |
| num_classes: null | |
| test_ds: | |
| prompt_template: '<extra_id_0>System | |
| {system message} | |
| <extra_id_1>User | |
| {turn 1 user message} | |
| <extra_id_1>Assistant | |
| <extra_id_2>{turn 1 assistant label} | |
| {turn 1 assistant message} | |
| <extra_id_1>User | |
| {turn 2 user message} | |
| <extra_id_1>Assistant | |
| <extra_id_2>{turn 2 assistant label} | |
| {turn 2 assistant message} | |
| <extra_id_1>' | |
| data_impl: jsonl | |
| splits_string: null | |
| seq_length: 8192 | |
| skip_warmup: true | |
| reset_position_ids: false | |
| reset_attention_mask: false | |
| eod_mask_loss: false | |
| index_mapping_dir: null | |
| data_prefix: | |
| train: | |
| - /dataset/train.jsonl | |
| validation: | |
| - /dataset/val.jsonl | |
| test: | |
| - /dataset/val.jsonl | |
| nsys_profile: | |
| enabled: false | |
| start_step: 10 | |
| end_step: 10 | |
| ranks: | |
| - 0 | |
| gen_shape: false | |
| optim: | |
| name: distributed_fused_adam | |
| lr: 2.0e-07 | |
| weight_decay: 0.1 | |
| betas: | |
| - 0.9 | |
| - 0.98 | |
| sched: | |
| name: CosineAnnealing | |
| warmup_steps: 10 | |
| constant_steps: 1000 | |
| min_lr: 1.9999e-07 | |
| max_steps: 88 | |
| bucket_cap_mb: 200 | |
| overlap_grad_sync: false | |
| contiguous_grad_buffer: true | |
| rotary_base: 500000.0 | |
| precision: bf16 | |
| answer_only_loss: true | |
| restore_from_path: /models/unpacked_llama3_70b_base | |
| save_nemo_on_validation_end: true | |
| use_flash_attention: null | |
| pipeline_model_parallel_split_rank: 0 | |
| ppo: | |
| trt_llm: | |
| enable: true | |
| reshard: true | |
| max_input_len: 4096 | |
| max_input_tokens: 32768 | |
| model_type: LLaMAForCausalLM | |
| unload_engine_train: false | |
| rollout_micro_batch_size: 8 | |
| num_rollout_samples: 128 | |
| forward_micro_batch_size: 4 | |
| val_rollout_micro_batch_size: 8 | |
| num_val_samples: 128 | |
| offload_adam_states: true | |
| entropy_bonus: 0.0 | |
| ratio_eps: 0.2 | |
| sampling_params: | |
| use_greedy: false | |
| temperature: 1.0 | |
| top_k: 0 | |
| top_p: 1.0 | |
| repetition_penalty: 1.0 | |
| add_BOS: false | |
| all_probs: false | |
| compute_logprob: false | |
| end_strings: | |
| - <|endoftext|> | |
| - <extra_id_1> | |
| length_params: | |
| max_length: 2048 | |
| min_length: 1 | |
| peft: | |
| peft_scheme: none | |
| restore_from_path: null | |
| restore_from_ckpt: | |
| checkpoint_dir: null | |
| checkpoint_name: null | |
| lora_tuning: | |
| target_modules: | |
| - attention_qkv | |
| adapter_dim: 32 | |
| adapter_dropout: 0.0 | |
| column_init_method: xavier | |
| row_init_method: zero | |
| layer_selection: null | |
| weight_tying: false | |
| position_embedding_strategy: null | |
| offload_adam_states: true | |
| enable_nge: true | |
| target: nemo_aligner.models.nlp.gpt.megatron_gpt_ppo_actor.MegatronGPTActorModel | |
| nemo_version: 2.0.0rc0 | |