config.json

{
  "source_checkpoint_size_gb": 15.266361777,
  "num_kept_params": 116,
  "num_total_params": 455,
  "kept_size_mb": 34.873348,
  "extracted_with": "extract_lora_only.py",
  "parameter_prefixes_kept": [
    "lora_A",
    "lora_B",
    "reward_head",
    "step_proj",
    "step_embed"
  ],
  "training_config": {
    "batch_size": 4,
    "grad_accum": 8,
    "lr": 1e-05,
    "seed": 43,
    "lora_r": 16,
    "lora_alpha": 32,
    "lora_dropout": 0.05,
    "step_embed_dim": 256,
    "reward_hidden": 1024,
    "min_mask_ratio": 0.0,
    "max_mask_ratio": 0.0,
    "causal": true,
    "no_step_embed": true,
    "no_mask_aware": false,
    "pool_strategy": "last_token",
    "max_steps": 15000
  }
}