{
  "model_name": "gpt2small-en-it-nanochat-lr2e4-bs6-wsd-fastdecay-step10000",
  "selected_checkpoint": "step_10000.pt",
  "selection_reason": "best_validation.json minimum validation loss for this run",
  "best_validation": {
    "step": 10000,
    "validation_loss": 3.8945770747959614,
    "validation_perplexity": 49.1352684243327,
    "validation_num_batches": 128,
    "elapsed_sec": 82998.0124297142
  },
  "final_validation_step_10000": {
    "step": 10000,
    "validation_loss": 3.8945770747959614,
    "validation_perplexity": 49.1352684243327,
    "validation_num_batches": 128,
    "elapsed_sec": 82998.0124297142
  },
  "notes": [
    "This is the best saved checkpoint of the stable-config-recipe-v5-gpt2small-lr2e4-batchmaxpossible-bs6-wsd-fastdecay run.",
    "For this run the final saved checkpoint step_10000.pt is also the best validation checkpoint.",
    "Probe telemetry exists, but this run wrote legacy/null probe target fields, so probe quality claims are intentionally conservative."
  ],
  "tokenizer_dir": "/mnt/apps/llm-nanochat/tokenizers/tok_202605011052_fresh_50_50_score100_32k_fromscratch",
  "dataset_dir": "/mnt/apps/llm-nanochat/datasets/202605011052_fresh_50_50_score100_2500_sourcebalanced"
}