training_config.json

{
  "model_config": {
    "model_class": "LlamaForCausalLM",
    "vocab_size": 4096,
    "hidden_act": "silu",
    "max_position_embeddings": 512,
    "initializer_range": 0.02,
    "rms_norm_eps": 1e-06,
    "bos_token_id": 0,
    "eos_token_id": 1,
    "pad_token_id": 2,
    "tie_word_embeddings": false,
    "rope_theta": 10000.0,
    "rope_scaling": null,
    "attention_bias": false,
    "attention_dropout": 0.0,
    "hidden_size": 12,
    "intermediate_size": 48,
    "num_attention_heads": 2,
    "num_hidden_layers": 1,
    "num_key_value_heads": 1
  },
  "max_seq_len": 512,
  "run_name": "2024_05_25_09_15_16",
  "out_dir": "/home/paperspace/.local/share/delphi/2024_05_25_09_15_16",
  "device": "auto",
  "checkpoint_interval": 400,
  "extra_checkpoint_iters": [
    1,
    2,
    4,
    8,
    16,
    32,
    64,
    128,
    256,
    512
  ],
  "log_interval": 40,
  "eval_iters": 10,
  "resume_from_path": null,
  "batch_size": 256,
  "max_epochs": 1,
  "grad_clip": 1.0,
  "gradient_accumulation_steps": 4,
  "adam": {
    "learning_rate": 0.0005,
    "weight_decay": 0.1,
    "beta1": 0.9,
    "beta2": 0.95,
    "decay_lr": true,
    "warmup_iters": 1000,
    "min_lr": 0.0
  },
  "batch_ordering_seed": 1337,
  "torch_seed": 42,
  "save_optimizer": true,
  "dataset": {
    "path": "delphi-suite/stories-tokenized",
    "feature": "tokens",
    "train_split": "train[:3%]",
    "validation_split": "validation"
  },
  "tokenizer": "delphi-suite/stories-tokenizer",
  "wandb": "jettjaniak/delphi",
  "out_repo": "jettjaniak/stories-llama2-100k-v02-test3",
  "debug_config": {
    "no_training": false,
    "no_eval": false
  }
}