Lora finetuning on Wikipedia-10, applying increased dropout for all layers

Dataset: Wikipedia-10
Target modules = ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"]

{
    "epoch": 0.1857053320643469,
    "total_flos": 2.605973017460736e+18,
    "train_loss": 1.8399735396504402,
    "train_runtime": 135360.4815,
    "train_samples": 344632,
    "train_samples_per_second": 0.473,
    "train_steps_per_second": 0.015
}

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

Downloads last month: -

Safetensors

Model size

7B params

Tensor type

BF16