zlei6/ComfyUI
1
0
Fork 0
mirror of https://github.com/Comfy-Org/ComfyUI.git synced 2026-08-13 05:13:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix some clip vision regression. (#15506)

Browse Source
This commit is contained in:
comfyanonymous
2026-08-11 12:54:15 -07:00
committed by GitHub
parent 2eaf09f50d
commit d9f9d2ba12
1 changed files with 6 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
comfy/clip_model.py
View File

@@ -314,13 +314,18 @@ class CLIPVisionModelProjection(torch.nn.Module):
if "projection_dim" in config_dict:
self.visual_projection = operations.Linear(config_dict["hidden_size"], config_dict["projection_dim"], bias=False)
else:
self.visual_projection = lambda a: a
self.visual_projection = torch.nn.Identity()
if "llava3" == config_dict.get("projector_type", None):
self.multi_modal_projector = LlavaProjector(config_dict["hidden_size"], 4096, dtype, device, operations)
else:
self.multi_modal_projector = None
def _load_from_state_dict(self, state_dict, prefix, *args, **kwargs):
if "{}visual_projection.weight".format(prefix) not in state_dict:
self.visual_projection = torch.nn.Identity()
super()._load_from_state_dict(state_dict, prefix, *args, **kwargs)
def forward(self, *args, **kwargs):
x = self.vision_model(*args, **kwargs)
out = self.visual_projection(x[2])