14xp/

gpt2_ablation_data

This dataset contains gradient magnitude data from GPT-2 model ablation experiments, specifically focusing on different gradient computation methods and block configurations.

The dataset contains two main directories:

Contains gradient magnitude data using staircase gradient computation method:

Method: Staircase gradient computation Block Configuration: 3P (3 parallel blocks)… See the full description on the dataset page: https://huggingface.co/datasets/14xp/gpt2_ablation_data.