RuntimeError: Input type (torch.FloatTensor) and weight type (torch.HalfTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

wallish77 commented 12 months ago

Attempting to use IPAdapter for SD 1.5 models. Loaded previously-working workflow, error occurs on first IPAdapter node. Deleted and recreated IPAdapter node but still seeing error.

Using IPAdapater commit bbf9336c5c75caf7dd307298fa8c1fb4931180dc
Using ComfyUI commit comfyanonymous/ComfyUI@65cae62c7149048951fc70d601fc9c0d5434f69c

May be related to changes made in ComfyUI to CLIP Vision.

comfyanonymous/ComfyUI@4e89b2c25a2ee7346b4892120176fadb0e7f931a
comfyanonymous/ComfyUI@a094b45c93e29193ba2695e834aefe7bf4635c06

Arron17 commented 12 months ago

Same issue for me, I agree with wallish hypothosis. There have been changes made to how the Clip Vision module is working.

laksjdjf commented 12 months ago

Thanks for the report! I've tried a fix, but it caused problems with previous versions, so I've put it on the dev branch for now.

Arron17 commented 12 months ago

Thanks, that Dev version fixes the errors. Performance is better for me as well with the combination of this and the comfyui update.

laksjdjf / IPAdapter-ComfyUI

RuntimeError: Input type (torch.FloatTensor) and weight type (torch.HalfTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor #17