Updated conda environment to be able to run H100s - updated installation instructions in main README. Can still use iti_a100.yaml for previous version of packages
Hyperparameter swept ITI on Llama2 and Llama3 models; documented results of sweep on Llama3_80B documented in validation/llama3_70B_tuning.md
Added "instruction_prompt" hyperparameter to validate_2fold, which lets the user customize the prefix used whenever the model is being prompted to answer a question from TruthfulQA. Users can play around with different instruction_prompts with llama3_70b_inference.ipynb.
Added finetune_gpt.ipynb to streamline process of finetuning custom GPT-Judge and GPT-Info
Added playscript_benchmark directory; currently contains a Python notebook with ~100 premises for benchmarking on future ITI experiments; will be updated in the future
Main contributions: