Skip to content
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

[core] Fix peft multi-gpu issue #145

Merged
merged 7 commits into from
Mar 14, 2023
Merged

[core] Fix peft multi-gpu issue #145

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
9ea7519
add multi-gpu support
younesbelkada Mar 2, 2023
47acddd
rm deepcopy
younesbelkada Mar 2, 2023
6f40411
Merge remote-tracking branch 'upstream/main' into temp-2
younesbelkada Mar 10, 2023
be12bac
tryo to comment
younesbelkada Mar 14, 2023
a6b3968
Merge remote-tracking branch 'upstream/main' into temp-2
younesbelkada Mar 14, 2023
ae7c42f
style
younesbelkada Mar 14, 2023
01cfe6b
fix nits
younesbelkada Mar 14, 2023
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
6 changes: 6 additions & 0 deletions src/peft/tuners/lora.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -200,6 +200,11 @@ def _replace_module(self, parent_module, child_name, new_module, old_module):
new_module.state = old_module.state
new_module.to(old_module.weight.device)

# dispatch to correct device
for name, module in new_module.named_modules():
if "lora_" in name:
module.to(old_module.weight.device)

def __getattr__(self, name: str):
"""Forward missing attributes to the wrapped module."""
try:
Expand Down Expand Up @@ -345,6 +350,7 @@ def forward(self, x: torch.Tensor):
transpose(self.lora_B.weight @ self.lora_A.weight, self.fan_in_fan_out) * self.scaling
)
self.merged = False

return F.linear(x, transpose(self.weight, self.fan_in_fan_out), bias=self.bias)
elif self.r > 0 and not self.merged:
result = F.linear(x, transpose(self.weight, self.fan_in_fan_out), bias=self.bias)
Expand Down