Skip to content

[Cherry-Pick][RL]Fix RL weight loading issue in moe layer #5503 #5505

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
Jiang-Jia-Jun merged 1 commit into from
Dec 11, 2025
Merged

[Cherry-Pick][RL]Fix RL weight loading issue in moe layer #5503 #5505

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
27 changes: 3 additions & 24 deletions fastdeploy/model_executor/layers/moe/fused_moe_backend_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,9 +23,8 @@
from fastdeploy.model_executor.utils import (
TensorTracker,
default_weight_loader,
free_tensor,
process_weight_transpose,
set_weight_attrs,
weight_fully_copied,
)
from fastdeploy.platforms import current_platform

Expand Down Expand Up @@ -312,25 +311,5 @@ def create_weights(self, layer: nn.Layer, **extra_weight_attrs):
def process_weights_after_loading(self, layer):
if self.model_format != "torch":
return
if not weight_fully_copied(layer.up_gate_proj_weight) or not weight_fully_copied(layer.down_proj_weight):
return
up_gate_proj_weight_transpose = layer.up_gate_proj_weight.transpose([0, 2, 1])
down_proj_weight_transpose = layer.down_proj_weight.transpose([0, 2, 1])
up_gate_proj = layer.create_parameter(
shape=up_gate_proj_weight_transpose.shape,
dtype=up_gate_proj_weight_transpose.dtype,
default_initializer=paddle.nn.initializer.Normal(mean=0.0, std=0.02),
is_bias=False,
)
up_gate_proj.copy_(up_gate_proj_weight_transpose, False)
free_tensor(layer.up_gate_proj_weight)
layer.up_gate_proj_weight = up_gate_proj
down_proj = layer.create_parameter(
shape=down_proj_weight_transpose.shape,
dtype=down_proj_weight_transpose.dtype,
default_initializer=paddle.nn.initializer.Normal(mean=0.0, std=0.02),
is_bias=False,
)
down_proj.copy_(down_proj_weight_transpose, False)
free_tensor(layer.down_proj_weight)
layer.down_proj_weight = down_proj
process_weight_transpose(layer, "up_gate_proj_weight")
process_weight_transpose(layer, "down_proj_weight")
Loading