This repository provides training scripts for fine-tuning Qwen2-VL, Qwen2.5-VL, Qwen3-VL, and Qwen3.5 vision-language models. It supports supervised fine-tuning (SFT), DPO, and GRPO training methods, with options for video training, reasoning mode training, and MoE model support. The implementation uses HuggingFace transformers and Liger-Kernel for efficient training.