Untitled

checkpoints:
  checkpoint_interval: 1250
  checkpoints_path: /mbz/users/fan.zhou/backup/storage/finemath/pt/llama_3_1_8B_nanotron_cpt/megamath/mixture_bs4M_constant_lr1e-5_seq8k_200B_dp16_tp8
  checkpoints_path_is_shared_file_system: false
  load_lr_scheduler: true
  load_optimizer: true
  resume_checkpoint_path: /mbz/users/fan.zhou/backup/storage/finemath/pt/llama_3_1_8B_nanotron_cpt/megamath/mixture_bs4M_constant_lr1e-5_seq8k_200B_dp16_tp8/30000
  save_final_state: true
  save_initial_state: false
data_stages:
- data:
    dataset:
      dataset_folder:
      - /mbz/users/fan.zhou/backup/storage/finemath/tokenized_data/llama3/dclm
      - /mbz/users/fan.zhou/backup/storage/finemath/tokenized_data/llama3/megamath_web_pro++
      - /mbz/users/fan.zhou/backup/storage/finemath/tokenized_data/llama3/megamath/megamath_code
      - /mbz/users/fan.zhou/backup/storage/finemath/tokenized_data/llama3/megamath/megamath_synth_qa
      - /mbz/users/fan.zhou/backup/storage/finemath/tokenized_data/llama3/megamath/megamath_synth_code
      - /mbz/users/fan.zhou/backup/storage/finemath/tokenized_data/llama3/megamath/megamath_synth_code_block
      dataset_weights:
      - 0.1
      - 0.725
      - 0.0125
      - 0.05
      - 0.0125
      - 0.1
    num_loading_workers: 1
    seed: 42
  name: Stable Training Stage
  start_training_step: 1
general:
  benchmark_csv_path: null
  consumed_train_samples: null
  ignore_sanity_checks: true
  project: MATH-BASE-CPT
  run: llama_3_1_8B_megamath_mixture_bs8M_constant_lr1e-5_seq8k_200B_dp16_tp8
  seed: 42
  id: xrcvmbh8
  step: null
lighteval: null
logging:
  iteration_step_info_interval: 1
  log_level: info
  log_level_replica: info
model:
  ddp_bucket_cap_mb: 25
  dtype: bfloat16
  init_method:
    path: /mbz/users/fan.zhou/backup/storage/ckpts/Llama-3.1-8B-nanotron/
  make_vocab_size_divisible_by: 1
  model_config:
    bos_token_id: 128000
    eos_token_id: 128001
    hidden_act: silu
    hidden_size: 4096
    initializer_range: 0.02
    intermediate_size: 14336
    is_llama_config: true
    max_position_embeddings: 131072
    num_attention_heads: 32
    num_hidden_layers: 32
    num_key_value_heads: 8
    pad_token_id: null
    pretraining_tp: 1
    rms_norm_eps: 1.0e-05
    rope_interleaved: false
    rope_scaling:
      factor: 8.0
      high_freq_factor: 4.0
      low_freq_factor: 1.0
      original_max_position_embeddings: 8192
      rope_type: llama3
    rope_theta: 500000.0
    tie_word_embeddings: false
    use_cache: true
    vocab_size: 128256
optimizer:
  accumulate_grad_in_fp32: true
  clip_grad: 1.0
  learning_rate_scheduler:
    learning_rate: 1.0e-05
    lr_decay_starting_step: null
    lr_decay_steps: 50000
    lr_decay_style: linear
    lr_warmup_steps: 0
    lr_warmup_style: linear
    min_decay_lr: 1.0e-05
  optimizer_factory:
    adam_beta1: 0.9
    adam_beta2: 0.95
    adam_eps: 1.0e-08
    name: adamW
    torch_adam_is_fused: true
  weight_decay: 0.01
  zero_stage: 0
parallelism:
  dp: 32
  expert_parallel_size: 1
  pp: 1
  pp_engine: 1f1b
  recompute_layer: false
  tp: 8
  tp_linear_async_communication: true
  tp_mode: REDUCE_SCATTER
  tp_recompute_allgather: true
profiler: null
s3_upload: null
tokenizer:
  tokenizer_max_length: null
  tokenizer_name_or_path: /mbz/users/fan.zhou/backup/storage/ckpts/Llama-3.1-8B-nanotron/
  tokenizer_revision: null
tokens:
  batch_accumulation_per_replica: 4
  limit_test_batches: 0
  limit_val_batches: 0
  micro_batch_size: 4
  sequence_length: 8192
  train_steps: 50000
  val_check_interval: -1