jinx/ComfyUI
1
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-08-02 23:14:49 +08:00
Code Issues Activity

Properly set if clip text pooled projection instead of using hack.

Browse Source
This commit is contained in:
comfyanonymous
2024-08-20 10:00:16 -04:00
parent 538cb068bc
commit 83dbac28eb
4 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
comfy/sd1_clip.py
View File

@@ -75,7 +75,6 @@ class ClipTokenWeightEncoder:
return r
class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
"""Uses the CLIP transformer encoder for text (from huggingface)"""
LAYERS = [
"last",
"pooled",
@@ -556,8 +555,12 @@ class SD1Tokenizer:
def state_dict(self):
return {}
class SD1CheckpointClipModel(SDClipModel):
def __init__(self, device="cpu", dtype=None, model_options={}):
super().__init__(device=device, return_projected_pooled=False, dtype=dtype, model_options=model_options)
class SD1ClipModel(torch.nn.Module):
def __init__(self, device="cpu", dtype=None, model_options={}, clip_name="l", clip_model=SDClipModel, name=None, **kwargs):
def __init__(self, device="cpu", dtype=None, model_options={}, clip_name="l", clip_model=SD1CheckpointClipModel, name=None, **kwargs):
super().__init__()
if name is not None: