gpt2-cda-race / README.md
ao9000's picture
Update README.md
db698ff verified

Lora finetuning on Wikipedia-10, applying counter factual data augmentation (CDA)

  • Dataset: Wikipedia-10
  • Target modules = ["query", "value"]
{
    "epoch": 2.987443401951174,
    "total_flos": 1.04520375926784e+18,
    "train_loss": 0.8730267834961414,
    "train_runtime": 69929.3152,
    "train_samples": 21423,
    "train_samples_per_second": 0.915,
    "train_steps_per_second": 0.029
}

Training script: https://github.com/ao9000/bias-bench/blob/main/experiments/run_clm.py