Skip to content

[LLM] disable part of MC2 in lora #8505

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
May 29, 2024
Merged

[LLM] disable part of MC2 in lora #8505

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
6 changes: 4 additions & 2 deletions paddlenlp/peft/lora/lora_layers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -409,7 +409,8 @@ def forward(self, x: paddle.Tensor):

if not self.merged:
input_mp = self.lora_dropout(input_mp)
if MC2RowSeqParallelCoreLinear is None:
# TODO(@gexiao): temporary workaround for deterministic calculation
if True or MC2RowSeqParallelCoreLinear is None:
input_mp = input_mp @ self.lora_A
input_mp = ReduceScatterOp.apply(input_mp)
else:
Expand Down Expand Up @@ -651,7 +652,8 @@ def forward(self, x: paddle.Tensor):

if not self.merged:
input_a = self.lora_dropout(x) @ self.lora_A
if MC2ColumnSeqParallelCoreLinear is None:
# TODO(@gexiao): temporary workaround for deterministic calculation
if True or MC2ColumnSeqParallelCoreLinear is None:
input_a = AllGatherOp.apply(input_a)
delta_mp = (input_a @ self.lora_B) * self.scaling
else:
Expand Down