{
  "schema_version": 1,
  "execution_location": "kaggle-free",
  "external_service_spend_usd": 0,
  "summary": {
    "run_id": "experiment-01",
    "state": "success",
    "hardware": {
      "cuda": "12.8",
      "device": "cuda",
      "name": "Tesla T4"
    },
    "global_step": 100,
    "trainable_parameters": 460800,
    "base_f1": 0.0,
    "trained_f1": 0.04899905814302255,
    "base_invalid_rate": 1.0,
    "trained_invalid_rate": 0.9564369310793238,
    "test_count": 3076,
    "improved": 96,
    "regressed": 0,
    "peak_allocated_bytes": 616607744,
    "peak_reserved_bytes": 675282944,
    "elapsed_seconds": 3412.8421733860005,
    "setup_seconds": 5.540617005000058,
    "steady_step_seconds_median": 0.501931781000053,
    "steady_examples_per_second": 15.752315262006347,
    "steady_tokens_per_second": 1162.4565711717337,
    "final_attempt_elapsed_seconds": 712.5247101720001,
    "attempts": [
      {
        "attempt": 0,
        "elapsed_seconds": 2700.3174632140003,
        "error": "Hard wall-clock deadline exceeded",
        "restored_step": 0,
        "started_at": "2026-09-17T15:08:52.821746+00:00",
        "state": "failed"
      },
      {
        "attempt": 1,
        "elapsed_seconds": 712.5247101720001,
        "restored_step": 100,
        "started_at": "2026-09-17T15:56:23.961623+00:00",
        "state": "success"
      }
    ],
    "generation_batch_size": 8,
    "peak_memory_scope": "Successful evaluation resume only; hard-killed training attempt did not finalize peak memory"
  },
  "recovery": {
    "physical_gpu_index": 1,
    "visible_device": "Tesla T4",
    "scope": "single-GPU LoRA recovery, not DDP",
    "evaluation_concurrently_on_gpu_index": 0,
    "run_ids": [
      "gpu-recovery-reference",
      "gpu-recovery-resumed"
    ],
    "final_step": 4,
    "restored_steps": [
      0,
      2
    ],
    "fault": {
      "signal": "SIGKILL",
      "step": 3
    },
    "all_checkpoint_state_matches": true,
    "max_abs_tensor_difference": 0.0,
    "rtol": 1e-05,
    "atol": 1e-06
  },
  "local_import": {
    "validation_location": "local-macos",
    "training_location": "kaggle-free",
    "first_import": {
      "lora_runs": 4,
      "lora_metrics": 111,
      "lora_predictions": 3076
    },
    "second_import": {
      "lora_runs": 4,
      "lora_metrics": 111,
      "lora_predictions": 3076
    },
    "quality_evaluation": true,
    "archive_sha256": "635e3c2b145de088b365fc5aa9c80cde4f2c1cd404d0b35f48c3e77da52d1080"
  },
  "checkpoint_download": {
    "archive_sha256": "6dc8a3d03bb045a31bed0a424cfa6a933f2ba8842b15f7069f01b2d41212fb2c",
    "archive_bytes": 81820458,
    "validated_files": 58,
    "final_checkpoint_steps": {
      "experiment-01": 100,
      "gpu-recovery-reference": 4,
      "gpu-recovery-resumed": 4
    },
    "validation": "Every extracted checkpoint and adapter file matches its downloaded run manifest hash"
  },
  "databricks": {
    "state": "externally-verified",
    "execution_location": "databricks-free-edition-reporting",
    "training_location": "kaggle-free",
    "run_ids": [
      "experiment-01",
      "gpu-recovery-reference",
      "gpu-recovery-resumed",
      "smoke"
    ],
    "first_import": {
      "lora_runs": 4,
      "lora_metrics": 111,
      "lora_predictions": 3076
    },
    "second_import": {
      "lora_runs": 4,
      "lora_metrics": 111,
      "lora_predictions": 3076
    },
    "payload_row_counts": {
      "lora_runs": 4,
      "lora_metrics": 111,
      "lora_predictions": 3076
    },
    "canonical_metrics": 110,
    "duplicate_key_groups": 0,
    "spark": "4.2.0",
    "python": "3.12.3 (main, Jun 19 2026, 12:46:00) [GCC 13.3.0]",
    "source_payload_sha256": "22f295e057b866b88aff8f8cd87b8b6d62a7325ea1bcc14c14f6b3076f52216c",
    "external_service_spend_usd": 0,
    "capture": {
      "method": "Transcribed JSON output from the executed notebook via Codex browser accessibility tree",
      "date_utc": "2026-09-17",
      "notebook_url": "https://dbc-d5f3aa4b-820e.cloud.databricks.com/editor/notebooks/1387683565256667?o=7474645850009944",
      "all_four_code_cells_succeeded": true,
      "compute": "Existing Free Edition CPU Serverless"
    }
  },
  "model": {
    "id": "HuggingFaceTB/SmolLM2-135M-Instruct",
    "revision": "12fd25f77366fa6b3b4b768ec3050bf629380bac",
    "license": "Apache-2.0"
  },
  "dataset": {
    "id": "mteb/banking77",
    "revision": "18072d2685ea682290f7b8924d94c62acc19c0b2",
    "content_sha256": "375427283fcef7a1e557ab1f9626fb731f7543419b2571e53d055017669dcb07",
    "attribution": "BANKING77: Casanueva, Temcinas, Gerz, Henderson and Vulic (2020), PolyAI; Hugging Face mirror by MTEB",
    "license": "CC-BY-4.0 (upstream); MIT (mirror card)"
  },
  "limitations": [
    "95.64% invalid adapter labels; not a usable banking classifier",
    "Single-GPU jobs, not GPU DDP",
    "First attempt hit its 45-minute deadline; successful resume did not repeat training",
    "Successful resume memory peaks exclude the hard-killed training attempt",
    "No repeated controlled GPU performance benchmark",
    "No host-loss recovery or production claim"
  ]
}