Can this model be used in diffusers?
#6
by yxliu - opened
I tried to load the fp8 safetensors with QwenImageTransformer2DModel.from_pretrained, and it says all the weight_scale tensors are not used. Is it loading correctly, or is it wrong as these tensors are necessary for inference? If this is not the correct way to load it, are there any way to use this model in diffusers? Thanks.
you need non-scaled FP8 version with LoRA, this is only for LightX2V and ComfyUI
Thanks