Skip to content

Fix bug with applying loras on fp8 scaled without fp8 ops. #10279

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 9, 2025
Merged

Fix bug with applying loras on fp8 scaled without fp8 ops. #10279

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 8 additions & 4 deletions comfy/model_patcher.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,17 +130,21 @@ def __init__(self, key, patches, convert_func=None, set_func=None):
self.set_func = set_func

def __call__(self, weight):
intermediate_dtype = weight.dtype
if self.convert_func is not None:
weight = self.convert_func(weight.to(dtype=torch.float32, copy=True), inplace=True)

intermediate_dtype = weight.dtype
if self.set_func is None and intermediate_dtype not in [torch.float32, torch.float16, torch.bfloat16]: #intermediate_dtype has to be one that is supported in math ops
if intermediate_dtype not in [torch.float32, torch.float16, torch.bfloat16]: #intermediate_dtype has to be one that is supported in math ops
intermediate_dtype = torch.float32
return comfy.float.stochastic_rounding(comfy.lora.calculate_weight(self.patches[self.key], weight.to(intermediate_dtype), self.key, intermediate_dtype=intermediate_dtype), weight.dtype, seed=string_to_seed(self.key))
out = comfy.lora.calculate_weight(self.patches[self.key], weight.to(intermediate_dtype), self.key, intermediate_dtype=intermediate_dtype)
if self.set_func is None:
return comfy.float.stochastic_rounding(out, weight.dtype, seed=string_to_seed(self.key))
else:
return self.set_func(out, seed=string_to_seed(self.key), return_weight=True)

out = comfy.lora.calculate_weight(self.patches[self.key], weight, self.key, intermediate_dtype=intermediate_dtype)
if self.set_func is not None:
return self.set_func(out, seed=string_to_seed(self.key), return_weight=True)
return self.set_func(out, seed=string_to_seed(self.key), return_weight=True).to(dtype=intermediate_dtype)
else:
return out

Expand Down