training_metadata.json

{
  "base_model": "Qwen/Qwen2.5-7B-Instruct",
  "quantization": "4bit",
  "lora_config": {
    "r": 32,
    "alpha": 64,
    "dropout": 0.05,
    "target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "up_proj",
      "down_proj"
    ]
  },
  "training_config": {
    "epochs": 3,
    "batch_size": 2,
    "gradient_accumulation_steps": 8,
    "learning_rate": 0.0002,
    "lr_scheduler": "cosine",
    "warmup_ratio": 0.05,
    "max_seq_length": 2048,
    "bf16": true
  },
  "dataset_stats": {
    "train_examples": 37542,
    "val_examples": 4171
  },
  "results": {
    "train_loss": 0.25181230885952716,
    "eval_loss": 0.21678349375724792,
    "elapsed_seconds": 62600.10093379021,
    "peak_vram_gb": 23.689548015594482
  },
  "dry_run": false,
  "timestamp": "2026-03-21T21:06:36.174888+00:00"
}