{
  "stage": 4,
  "dataset_id": "THemidli/applied-ner-stage4-final",
  "base_model": "prajjwal1/bert-mini",
  "base_revision": "5e123abc2480f0c4b4cac186d3b3f09299c258fc",
  "tokenizer_model": "google/bert_uncased_L-2_H-128_A-2",
  "tokenizer_revision": "30b0a37ccaaa32f332884b96992754e246e48c5f",
  "seed": 20260802,
  "labels": [
    "O",
    "B-PERSON",
    "I-PERSON",
    "B-ORGANIZATION",
    "I-ORGANIZATION",
    "B-LOCATION",
    "I-LOCATION",
    "B-TIMEDATE",
    "I-TIMEDATE",
    "B-PRODUCT",
    "I-PRODUCT",
    "B-WORKOFART",
    "I-WORKOFART",
    "B-JOB",
    "I-JOB",
    "B-AMOUNT",
    "I-AMOUNT"
  ],
  "parameter_count": 11109137,
  "trainable_parameter_count": 11109137,
  "device": "mps",
  "platform": "macOS-27.0-arm64-arm-64bit",
  "torch_version": "2.13.0",
  "train_records": 841,
  "test_records": 159,
  "train_chunks": 865,
  "test_chunks": 165,
  "max_length": 256,
  "wall_seconds": 32.908,
  "trainer_metrics": {
    "train_runtime": 32.8519,
    "train_samples_per_second": 421.285,
    "train_steps_per_second": 13.637,
    "total_flos": 49534075078794.0,
    "train_loss": 0.23767771824662173,
    "epoch": 16.0
  },
  "hyperparameters": {
    "epochs": 16,
    "learning_rate": 0.0005,
    "train_batch_size": 32,
    "eval_batch_size": 64,
    "weight_decay": 0.02,
    "label_smoothing_factor": 0.0,
    "warmup_steps": 45,
    "scheduler": "linear",
    "hidden_dropout": 0.1,
    "attention_dropout": 0.1,
    "classifier_dropout": 0.1
  },
  "train_overall": {
    "overall_precision": 0.9988776655443322,
    "overall_recall": 0.998653500897666,
    "overall_f1": 0.9987655706430254,
    "overall_accuracy": 0.9998373013720918
  },
  "test_overall": {
    "overall_precision": 0.618925831202046,
    "overall_recall": 0.6747211895910781,
    "overall_f1": 0.6456202756780791,
    "overall_accuracy": 0.8791745256851214
  },
  "test_f1_change_vs_stage3": 0.1771147178353455,
  "mps_run_variance_note": "On the MPS backend, repeated runs with identical seed and config showed ±0.01–0.015 F1 variation (0.6331 vs 0.6455 across the two recorded runs); the seed is fixed, and the variation does not change the model ranking (both runs far above BERT-Tiny, slightly below ELECTRA-Small).",
  "dataset_revision": "ce231175a828d21865c638d9718c4a6e8ba1fb1c",
  "vocab_sha256": "07eced375cec144d27c900241f3e339478dec958f92fddbc551f295c992038a3"
}