Image-to-Image
Diffusers
English

Bug about ipadapter v2

#1
by YueMafighting - opened

RuntimeError: Error(s) in loading state_dict for ImageProjModel:
size mismatch for proj.weight: copying a param with shape torch.Size([65536, 768]) from checkpoint, the shape in current model is torch.Size([16384, 768]).
size mismatch for proj.bias: copying a param with shape torch.Size([65536]) from checkpoint, the shape in current model is torch.Size([16384]).

src/flux/xflux_pipeline.py, line 77:
self.improj = ImageProjModel(4096, 768, 4)

same error

I updated the Comfyui version to the latest and also updated the XLABS plugin to the latest, which solved the problem

XLabs AI org

please update comfy nodes, it'd solve the problem you mentioned

XLabs AI org

The problem says that model graph expect 4 visual tokens (was in previous version) now we have 16, update nodes to support both ip-adapters

XLabs AI org

RuntimeError: Error(s) in loading state_dict for ImageProjModel:
size mismatch for proj.weight: copying a param with shape torch.Size([65536, 768]) from checkpoint, the shape in current model is torch.Size([16384, 768]).
size mismatch for proj.bias: copying a param with shape torch.Size([65536]) from checkpoint, the shape in current model is torch.Size([16384]).

src/flux/xflux_pipeline.py, line 77:
self.improj = ImageProjModel(4096, 768, 4)

stazizov changed discussion status to closed

Sign up or log in to comment