yxf0314

This is an issue within SGLang diffusion itself, causing qwen-image-edit to incorrectly route through the video pipeline, resulting in missing fields. I tried launching SGLang Diffusion directly, without going through...

https://github.com/sgl-project/sglang/issues/13878 Based on the response to this issue, I pulled the latest version of SGLang and verified that it can now run Qwen-Image-Edit normally. Now we just need to wait...

https://github.com/sgl-project/sglang/commit/7ae368efde14f0b2abe769a5fbb99b8154b2a9b7 SGLang has been bumped to version 0.5.6, so we can schedule its integration into the runner. @thxCode @gitlawr

> 使用的是cosy2吗？俺也一样 是的，CosyVoice2-0.5B

> [生成音频.zip](https://github.com/user-attachments/files/21179535/default.zip)，我上传附件，你听听是不是这个情况 是的是的，像喝醉酒胡言乱语那样

@grvlbit Regarding the 2nd point, could you tell me what authentication method the issuer in your application uses? > 2. The [decode_and_validate_token()](https://github.com/gpustack/gpustack/blob/70575fb64796ddc1540b2be60d2590b2681b7598/gpustack/routes/auth.py#L30) function is failing with a warning: `WARNING -...