Pretrained Model

The pretrained Llama2 can be found via the following link: https://huggingface.co/meta-llama/Llama-2-7b

Model development

The script for LoRA fine-tuning based on the Llama2-7b can be found at ./Llama-main/train/sft/finetune_lora.sh, and the implementation details are located in ./Llama-main/train/sft/finetune_clm_lora.py. Fine-tuning with multiple GPUs on a single machine can be achieved by modifying the --include localhost:0 parameter in the script.

Model Testing

The ./Llama-main/inference/ directory contains scripts for model testing. These scripts enable the generation of report summaries from descriptive findings produced by different models.

Model Evaluation and Comparison

The ./Llama-main/metrics directory includes scripts for evaluating model performance and comparing various models. These scripts provide analyses of t-test and the calculation of F1 scores.