Elpapudex/AGnews-Full

u2t01-bert-agnews

Training Data

• Dataset: agnews
• Base Model: bert-base-uncased
• Hyperparameters:
  • head_lr: 1e-3
  • body_lr: 2e-5
  • seed: 42

Metrics

{
  "loss": 0.3354416787624359,
  "accuracy": 0.9472368421052632,
  "macro_f1": 0.9473421817081283,
  "runtime": 31.8229,
  "samples_per_second": 238.822,
  "steps_per_second": 1.885,
  "epoch": 3.0,
  "method": "full",
  "trainable_params": 109485316,
  "total_train_seconds": 4504.269985198975
}

Headline Metric: Accuracy = 0.9472 | Macro F1 = 0.9473

Limitations

• English-only.
• Domain-bound (trained on news topics classification).

References

• Devlin et al., BERT, arXiv:1810.04805 §5.3
• Hugging Face Transformers Training Documentation