apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 17:17:14 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix ep lm head (#3244)

Browse Source 
Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com>
This commit is contained in:
RichardWooSJTU
2025-08-12 15:38:28 +08:00
committed by GitHub
parent f5164215be
commit 283da92bfa
1 changed files with 2 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
fastdeploy/model_executor/layers/lm_head.py
View File

@@ -118,9 +118,7 @@ class ParallelLMHead(nn.Layer):
if self.use_ep: if self.use_ep:
self.weight.set_value(get_tensor(state_dict.pop(self.weight_key)).astype(paddle.get_default_dtype())) self.weight.set_value(get_tensor(state_dict.pop(self.weight_key)).astype(paddle.get_default_dtype()))
if self.bias_key is not None: if self.bias_key is not None:
self.bias.set_value( self.bias.set_value(get_tensor(state_dict.pop(self.bias_key)).astype(paddle.get_default_dtype()))
get_tensor(state_dict.pop(self.linear_bias_key)).astype(paddle.get_default_dtype())
)
else: else:
if self.tie_word_embeddings: if self.tie_word_embeddings:
self.linear.weight.set_value( self.linear.weight.set_value(
@@ -148,7 +146,7 @@ class ParallelLMHead(nn.Layer):
""" """
logits = input logits = input
if self.use_ep: if self.use_ep:
if self.linear_bias_key is None: if self.bias_key is None:
logits = paddle.matmul(logits, self.weight) logits = paddle.matmul(logits, self.weight)
else: else:
logits = paddle.incubate.nn.functional.fused_linear(logits, self.weight, self.bias) logits = paddle.incubate.nn.functional.fused_linear(logits, self.weight, self.bias)