ao9000/phi2-cda-gender

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

• Dataset: Wikipedia-10
• Target modules = ["q_proj", "k_proj", "v_proj", "dense", "fc1", "fc2"]

{
    "epoch": 0.060952845355012227,
    "total_flos": 1.04520375926784e+18,
    "train_loss": 0.5059917987585068,
    "train_runtime": 72742.6851,
    "train_samples": 1049992,
    "train_samples_per_second": 0.88,
    "train_steps_per_second": 0.027
}

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