请问怎么多卡推理呢

[LHRYANG]

from hyimage.diffusion.pipelines.hunyuanimage_pipeline import HunyuanImagePipeline

pipe = HunyuanImagePipeline.from_pretrained(model_name=self.model_id, torch_dtype='bf16', device=device)

pipe.to(device)

这样好像不行

[KimbingNg]

Hi @LHRYANG, Inference parallelism is not implemented yet. If you have an OOM issue during inference, you can try using the fp8-quantized models.