metadata.json

{
  "base_model": "google/medgemma-4b-it",
  "base_adapter": "sciedoc/neurojenml",
  "adapter_type": "lora",
  "lora_config": {
    "r": 16,
    "alpha": 16,
    "dropout": 0,
    "target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "up_proj",
      "down_proj"
    ]
  },
  "training_config": {
    "epochs": 3,
    "learning_rate": 0.0002,
    "batch_size": 2,
    "gradient_accumulation": 4,
    "warmup_steps": 5,
    "fp16": false,
    "bf16": false,
    "max_seq_length": 2048,
    "optimizer": "adamw_8bit",
    "weight_decay": 0.01,
    "lr_scheduler": "cosine",
    "seed": 3407
  },
  "dataset": {
    "source": "training_data.jsonl",
    "total_examples": 3956,
    "train_examples": 1747,
    "val_examples": 195
  },
  "results": {
    "final_train_loss": 0.0878016290792138,
    "final_eval_loss": 0.059457048773765564,
    "training_time_seconds": 8703.843399763107,
    "quality_guardrail": "response_quality_score",
    "training_time_minutes": 145.1
  },
  "hardware": {
    "gpu": "Tesla T4",
    "gpu_count": 2
  },
  "hf_repo": "https://huggingface.co/sciedoc/neurojenml",
  "hf_pushed": true,
  "created_at": "2026-07-12T14:53:06.814047"
}