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": 2.8503986104306773,
    "total_flos": 1.0451807295707516e+18,
    "train_loss": 0.7933661967515946,
    "train_runtime": 65423.383,
    "train_samples": 22453,
    "train_samples_per_second": 0.978,
    "train_steps_per_second": 0.031
}

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

Downloads last month: 14

Safetensors

Model size

2.78B params

Tensor type

BF16

Inference Examples

Text Generation

Inference API (serverless) is not available, repository is disabled.