yxf0314

使用transformers==4.53.1版本，生成语音会混乱

20

使用transformers==4.53.1版本，生成语音会混乱，改用4.51.3版本则正常 以下是调用代码 ``` def speech( self, input: str, voice: Optional[str] = "Chinese Female", speed: float = 1, reponse_format: str = "mp3", **kwargs, ) -> str: if voice not in self._voices:...

fix: enhance backend parameter handling

1

https://github.com/gpustack/gpustack/issues/3537#issuecomment-3600862659 https://github.com/gpustack/gpustack/issues/3649

Devices with arch_family of Ascend910B cannot automatically select the corresponding vLLM image.

1

### GPUStack version v2.0.0 ### Operating System & CPU Architecture _No response_ ### GPU _No response_ ### ▶️ Steps to reproduce The user's Ascend 910B device is detected normally, but...

The evaluation of SGLang Diffusion should not be affected by mem-fraction-static.

1

### GPUStack version v2.0.0 ### Operating System & CPU Architecture _No response_ ### GPU _No response_ ### ▶️ Steps to reproduce During the evaluation of image models supported by SGLang...

https://github.com/gpustack/gpustack/issues/3534 1. Added the `_is_diffusion` attribute, incorporating `_is_diffusion` checks wherever `_mem_fraction_static` is referenced. 2. Introduced a weight calculation method for diffusion models that follows the pipeline orchestration defined in `model_index.json`....