This collection contains checkpoints and LoRA adapters for Llama-2-7B trained on the Python subset of StarCoder for up to 20 billion tokens.