This repo includes submission for HW2. The breakdown of this repo is as follow.
- trained_gpt2_hf/ : This directory includes the converted huggingface format of the trained model.
- diff-file.diff: This is the diff file between the submitted model and the baseline model.
- gpt2_train_Rope_RMSNorm.11671391.log: Training log.
- hellaswag_10shot.json: The evaluation results on the HellaSwag dataset I run by myself.
- model_00019560.bin: The trained model in .bin format.
For more details regarding the model, please refer to the report.
