This is finetune experiment on LLaMa3 for course DSAN5810. This respotory includes mainly 4 parts:
- datasets generation
- llama finetune training
- HELM
- LLaMA3 backend server
Please refer to the README.md in each folder for more details.
.
├── datasets # datasets generation
│ ├── README.md
│ ├── generate_datasets.py
├── helm # HELM evaluation
│ ├── README.md
│ ├── helm_commands.sh
│ ├── build_run_specs_full.py
│ ├── run_specs_full_coarse_600_budget.conf
├── llama_train # llama finetune training
│ ├── README.md
├── llama3_submission # LLaMA3 backend server
│ ├── api.py
│ ├── Dockerfile
│ ├── fast_api_requirements.txt
│ ├── main.py
│ └── start_server.sh
├── presentation # presentation slides
│ ├── imgs
│ ├── DSAN5810 Final Project.pdf
│ ├── DSAN5810 Final Project.pptx
├── README.md