unsloth vlm模型Qwen2-VL、Llama 3.2 Vision微调案例
T4卡15G显卡训练
参考：
https://github.com/unslothai/unsloth
按自己显卡cuda版本安装
免费colab微调代码：
Qwen2-VL：
https://colab.research.google.com/drive/1whHb54GNZMrNxIsi2wm2EY_-Pvo2QyKh?usp=sharing
from unsloth import FastVisionModel # NEW instead of FastLanguageModel
import torch
model, tokenizer = FastVisionModel.from_pretrained(
"unsloth/Q