This repository is a template for running experiments using the following tools and libraries:
- 📦 uv
- A Python package manager replacing
pip
andpoetry
.
- A Python package manager replacing
- 📝 pyproject
- For project management.
- ⚙️ hydra
- For flexible configuration management.
- 📊 wandb
- For experiment tracking and visualization online.
- 🤖 encourage
- A custom library for handling LLM inference, prompt handling, and utility functions.
To initialize the environment using uv
, run the following command:
uv sync
For using Weight & Biases you need to login to your account using the following command:
wandb login
There you have to enter your API key which you can find in your account settings.
When using this template you have to declare all your configuration parameters in the conf/defaults.yaml
file. Also modify the conf/model/defaults.yaml
and conf/data/defaults.yaml
files to fit your needs.
To run an experiment you can use the following command:
uv run main.py
The output will be saved in the outputs folder. Each experiment generates a new timestamped folder containing:
- Configuration files (.hydra/)
- Logs (main.log)
- Inference output (inference_log.json)