We use:
- hydra - configuration management
- ray - infrastructure management
- huggingface accelerate + deepspeed - enabling pytorch to run across any distributed compute configuration
- huggingface transformers - writing and managing models
Quickstart:
uv sync
uv run train.py