RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument tensors in method wrapper_CUDA_cat)

这个报错通常出现在 ComfyUI 使用 MiniMax H3 Video VAE 进行解码,且显存不足导致 VAE 模型被部分 CPU offload 时。优先排查 ComfyUI 是否已更新(Issue 末尾标记为已修复),并确认 `comfy/ldm/minimax/vae.py` 中相关参数

![InvokeError: [models] Error: 'required' occurs when invoking an Agent within a workflow (still exists in new version)](https://www.chat-gpts.plus/wp-content/uploads/2026/08/39937-cfeffb11-768x403.jpg)





![[Bug]: LiteLLM Cloudflare adapter doesn't translate OpenAI content-part messages](https://www.chat-gpts.plus/wp-content/uploads/2026/08/33984-0744c140-768x403.jpg)
