ComfyUI-IDM-VTON icon indicating copy to clipboard operation
ComfyUI-IDM-VTON copied to clipboard
verified publisher icon TemryL

It is recommended to add support for fp16 model to save memory

Open junne14105 opened this issue 1 year ago • 6 comments

https://huggingface.co/IDM-VTON-F16 The previous 12 g model memory requirements can be changed to 5 g model size. Will also allow and use better! What do you think?

junne14105 avatar May 03 '24 18:05 junne14105

hey! the link seems to be broken... the current implementation is already loading the model in torch.float16, I will have on how to load in 8bits

TemryL avatar May 03 '24 18:05 TemryL

Are you running on MPS accelerator?

TemryL avatar May 04 '24 10:05 TemryL

This VRAM usage info might be useful: image

From: https://github.com/yisol/IDM-VTON/issues/47

deepfree2023 avatar May 08 '24 05:05 deepfree2023