Skip to content

Fix dtype error in MHA layer/change dtype checking mechanism for manual cast #14791

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 4 commits into from
Jan 29, 2024
Merged

Fix dtype error in MHA layer/change dtype checking mechanism for manual cast #14791

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
750dd60
Fix potential bugs
KohakuBlueleaf Jan 29, 2024
6e7f086
linting
KohakuBlueleaf Jan 29, 2024
d243e24
Try to reverse the dtype checking mechanism
KohakuBlueleaf Jan 29, 2024
f9ba7e6
Revert "Try to reverse the dtype checking mechanism"
KohakuBlueleaf Jan 29, 2024
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
10 changes: 7 additions & 3 deletions modules/devices.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,6 @@

import torch
from modules import errors, shared
from modules import torch_utils

if sys.platform == "darwin":
from modules import mac_specific
Expand Down Expand Up @@ -141,7 +140,12 @@ def forward_wrapper(self, *args, **kwargs):
args = [arg.to(target_dtype) if isinstance(arg, torch.Tensor) else arg for arg in args]
kwargs = {k: v.to(target_dtype) if isinstance(v, torch.Tensor) else v for k, v in kwargs.items()}

org_dtype = torch_utils.get_param(self).dtype
org_dtype = target_dtype
for param in self.parameters():
if param.dtype != target_dtype:
org_dtype = param.dtype
break

if org_dtype != target_dtype:
self.to(target_dtype)
result = self.org_forward(*args, **kwargs)
Expand Down Expand Up @@ -170,7 +174,7 @@ def manual_cast(target_dtype):
continue
applied = True
org_forward = module_type.forward
if module_type == torch.nn.MultiheadAttention and has_xpu():
if module_type == torch.nn.MultiheadAttention:
module_type.forward = manual_cast_forward(torch.float32)
else:
module_type.forward = manual_cast_forward(target_dtype)
Expand Down