apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-04 00:06:38 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix bf16 and add comments (#4106)

Browse Source
This commit is contained in:
bukejiyu
2025-09-15 17:23:07 +08:00
committed by GitHub
parent 69aa2781a1
commit 113e330030
4 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
fastdeploy/model_executor/layers/linear.py
View File

@@ -498,6 +498,7 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
if weight_need_transpose:
loaded_weight = get_tensor(loaded_weight)
loaded_weight = loaded_weight.transpose([1, 0])
# Avoid redundant transpose of fused weights when weight_loader is called iteratively
param.weight_need_transpose = False
# Loaded weight is already fused on disk.
shard_offsets = [
@@ -638,6 +639,7 @@ class QKVParallelLinear(ColumnParallelLinear):
if weight_need_transpose:
loaded_weight = get_tensor(loaded_weight)
loaded_weight = loaded_weight.transpose([1, 0])
# Avoid redundant transpose of fused weights when weight_loader is called iteratively
param.weight_need_transpose = False
# Loaded weight is already fused on disk
shard_offsets = [