| checkpoints: | |
| checkpoint_interval: 1000 | |
| checkpoints_path: /fsx/phuc/checkpoints/doremi/big-run-02/reference-2.8b-llama | |
| checkpoints_path_is_shared_file_system: true | |
| resume_checkpoint_path: null | |
| save_initial_state: false | |
| data: | |
| dataset: | |
| dataset_overwrite_cache: false | |
| dataset_processing_num_proc_per_process: 1 | |
| hf_dataset_config_name: null | |
| hf_dataset_or_datasets: /fsx/phuc/project_data/doremi/datasets/the_pile_raw/tokenized_data/train | |
| hf_dataset_splits: train | |
| text_column_name: text | |
| num_loading_workers: 1 | |
| seed: 42 | |
| doremi: | |
| domain_names: | |
| - Pile-CC | |
| - Github | |
| - OpenWebText2 | |
| - StackExchange | |
| - Wikipedia (en) | |
| - PubMed Abstracts | |
| - USPTO Backgrounds | |
| - FreeLaw | |
| - PubMed Central | |
| - Enron Emails | |
| - HackerNews | |
| - NIH ExPorter | |
| - Books3 | |
| - ArXiv | |
| - DM Mathematics | |
| - OpenSubtitles | |
| - Gutenberg (PG-19) | |
| - Ubuntu IRC | |
| - BookCorpus2 | |
| - EuroParl | |
| - YoutubeSubtitles | |
| - PhilPapers | |
| domain_weights: null | |
| ref_model_checkpoint_path: null | |
| ref_model_resume_checkpoint_path: null | |
| general: | |
| benchmark_csv_path: null | |
| consumed_train_samples: 35840000 | |
| ignore_sanity_checks: true | |
| project: nanotron | |
| run: train_2.8b_llama_reference | |
| seed: 42 | |
| step: 70000 | |
| logging: | |
| iteration_step_info_interval: 1 | |
| log_level: info | |
| log_level_replica: info | |
| model: | |
| ddp_bucket_cap_mb: 120 | |
| dtype: bfloat16 | |
| init_method: | |
| std: 0.025 | |
| make_vocab_size_divisible_by: 1 | |
| model_config: | |
| bos_token_id: 1 | |
| eos_token_id: 2 | |
| hidden_act: silu | |
| hidden_size: 4096 | |
| initializer_range: 0.02 | |
| intermediate_size: 24576 | |
| is_llama_config: true | |
| max_position_embeddings: 1024 | |
| num_attention_heads: 32 | |
| num_hidden_layers: 6 | |
| num_key_value_heads: 16 | |
| pad_token_id: null | |
| pretraining_tp: 1 | |
| rms_norm_eps: 1.0e-05 | |
| rope_scaling: null | |
| tie_word_embeddings: true | |
| use_cache: true | |
| vocab_size: 49152 | |
| optimizer: | |
| accumulate_grad_in_fp32: true | |
| adam_beta1: 0.9 | |
| adam_beta2: 0.95 | |
| adam_eps: 1.0e-08 | |
| clip_grad: 1.0 | |
| learning_rate_scheduler: | |
| learning_rate: 0.0003 | |
| lr_decay_steps: 8 | |
| lr_decay_style: cosine | |
| lr_warmup_steps: 2 | |
| lr_warmup_style: linear | |
| min_decay_lr: 1.0e-05 | |
| torch_adam_is_fused: true | |
| weight_decay: 0.01 | |
| zero_stage: 0 | |
| parallelism: | |
| dp: 8 | |
| pp: 1 | |
| pp_engine: 1f1b | |
| recompute_granularity: SELECTIVE | |
| tp: 8 | |
| tp_linear_async_communication: true | |
| tp_mode: REDUCE_SCATTER | |
| profiler: null | |
| tokenizer: | |
| tokenizer_max_length: null | |
| tokenizer_name_or_path: gpt2 | |
| tokenizer_revision: null | |
| tokens: | |
| batch_accumulation_per_replica: 1 | |
| limit_test_batches: 0 | |
| limit_val_batches: 1 | |
| micro_batch_size: 64 | |
| sequence_length: 1024 | |
| train_steps: 70000 | |
| val_check_interval: 2 | |