Adapter/LoRA fine-tuning pipelines. Each subfolder is an independent uv
project (own pyproject.toml/.venv) — see its README for setup and data.
| Folder | Framework | Base model |
|---|---|---|
vicuna-7b-lora/ |
transformers + peft |
lmsys/vicuna-7b-v1.5, loaded directly (no LLaVA checkpoint, no vision encoder — see its README) |
qwen25-3b-lora/ |
transformers + peft |
Qwen/Qwen2.5-3B-Instruct — same pattern as vicuna-7b-lora/, smaller/faster, ChatML instead of Vicuna's USER:/ASSISTANT: (see its README) |