Qwen and DeepSeek low-VRAM LoRA training with GGUF base model, compatible with Transformers and PEFT - View it on GitHub