model_name: tangled-alpha-0.13-base
model_config:
  name: tangled-alpha-0.13-base
  hf_config: {}
  block_size: 131072
  n_layer: 32
  n_embd: 576
  vocab_size: 65536
  padding_multiple: 512
  padded_vocab_size: 65536
  norm_class_name: RMSNorm
  norm_eps: 1.0e-05
  norm_qk: false
  post_attention_norm: false
  post_mlp_norm: false
  parallel_residual: false
  shared_attention_norm: false
  n_head: 9
  head_size: 64
  n_query_groups: 3
  attn_bias: false
  rope_base: 84000
  rotary_percentage: 1.0
  rope_condense_ratio: 1
  intermediate_size: 1536
  bias: false
  mlp_class_name: LLaMAMLP
  gelu_approximate: none
  n_expert: 0
  n_expert_per_token: 0
  scale_embeddings: false
  lm_head_bias: false
out_dir: ../out/pretrain-base-0
precision: bf16-true
resume: auto
data:
  class_path: litgpt.data.LitData
  init_args:
    data_path: ../base-data-0-0-1073741824-8193-2000/
    seed: 42
    num_workers: 32
train:
  save_interval: 50
  log_interval: 1
  global_batch_size: 512
  micro_batch_size: 2
  lr_warmup_steps: 100
  max_tokens: 12261897783
  max_seq_length: 8193
  tie_embeddings: false
  max_norm: 1.0
  min_lr: 1.0e-05
eval:
  interval: 50
  max_iters: 100
  initial_validation: false
  final_validation: true
  evaluate_example: first
optimizer:
  class_path: sophia_opt.SophiaG
  init_args:
    lr: 0.001
    betas:
    - 0.965
    - 0.99
    rho: 0.04
    weight_decay: 0.1
devices: auto
num_nodes: 1
tokenizer_dir: ../tokenizer
logger_name: wandb
seed: 23