Getting fp16 model in diffusers?

#4
by chigozie - opened

If I do:

vae = ConsistencyDecoderVAE.from_pretrained(
    "openai/consistency-decoder",
    cache_dir='diffusers-cache',
    torch_dtype=torch.float16
)

as suggested in the model card, and then I go to look in diffusers-cache/snapshots/63b7a48896d92b6f56772f4111d0860b1bee3dd3, I see diffusion_pytorch_model.safetensors, not diffusion_pytorch_model.fp16.safetensors.

you need variant='fp16'

chigozie changed discussion status to closed

Sign up or log in to comment