chengzeyi / stable-fast

Best inference performance optimization framework for HuggingFace Diffusers on NVIDIA GPUs.
MIT License
1.19k stars 74 forks source link

Is stable fast support aingdiffusion_v12 ? #163

Open youthbupt opened 1 month ago

youthbupt commented 1 month ago

I tried aingdiffusion_v12 follows the code,but it didn't work (The time cost is the same comparing to not compile it).

model_path: models/stable_diffusion/aingdiffusion_v12.safetensors, modelname: sd_unet

The Code: config = CompilationConfig.Default() if config.memory_format is not None: self.unet = self.unet.to(memory_format=config.memory_format) self.unet = torch.compile(self.unet, backend=functools.partial( sfast_jit_trace, ts_compiler=_build_ts_compiler(config)))