2024 Final project for Deep Learning

This is the fork of LLamafactory used to perform all finetuning experiments.

Our custom YAML files used for fine-tuning can be found in project/<DATASET>/train. Our custom YAML files used for computing ROUGE scores are found in project/<DATASET>/eval. Our collected results and graphs are found in project/<DATASET>/results.

To run LLamafactory with our custom YAML files first clone this repository and run:

cd LLaMA-Factory
pip install -e .[torch,metrics]

Then run:

nvidia-smi --query-gpu=index,timestamp,utilization.gpu,memory.total,memory.free,memory.used --format=csv, --loop=10 > MEMORY_LOG_FILE_NAME.csv 2>&1 & CUDA_VISIBLE_DEVICES=0 llamafactory-cli train project/<DATASET>/train/llama3_<DATASET>_<SFT_METHOD_NAME>.yaml

To evaluate, run:

CUDA_VISIBLE_DEVICES=0 llamafactory-cli train project/<DATASET>/eval/llama3_<DATASET>_<SFT_METHOD_NAME>.yaml

The model and the results should be saved in the saves folder not tracked by this repository.