01-run-metadata.json

{
  "run_id": "sft-qwen2.5-7b-k8s-001",
  "objective": "supervised_fine_tuning",
  "base_model": {
    "name": "Qwen/Qwen2.5-7B",
    "revision": "main"
  },
  "dataset": {
    "name": "llm-k8s-sft",
    "version": "v1.3.2",
    "train_samples": 128400,
    "eval_samples": 3200,
    "estimated_train_tokens": 217843502
  },
  "peft": {
    "type": "lora",
    "r": 64,
    "alpha": 128,
    "target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "up_proj",
      "down_proj"
    ]
  },
  "runtime": {
    "nodes": 1,
    "gpus_per_node": 4,
    "precision": "bf16"
  },
  "artifacts": {
    "latest_checkpoint": "/mnt/models/runs/sft-qwen2.5-7b-k8s-001/checkpoint-000120",
    "adapter_dir": "/mnt/models/adapters/qwen2.5-7b-k8s-sft-v1",
    "merged_model_dir": "/mnt/models/merged/qwen2.5-7b-k8s-sft-v1"
  },
  "metrics": {
    "train_loss": 1.76,
    "eval_loss": 1.82,
    "domain_accuracy": 0.87,
    "safety_refusal_pass_rate": 0.98
  },
  "lineage": {
    "dataset_card": "s3://ml-assets/datasets/llm-k8s-sft/v1.3.2/dataset-card.yaml",
    "code_commit": "4b7e2c1",
    "container_image": "registry.local/llm/trainer:2026-04-10"
  }
}