test/fastNLP
1
0
Fork 0
mirror of https://gitee.com/fastnlp/fastNLP.git synced 2024-12-03 12:47:35 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

修改GradientClipCallback中parameter的存储方式，防止在torch 1.5版本中报错

Browse Source
This commit is contained in:
yh_cc 2020-04-29 00:13:13 +08:00
parent fbd2fd4ead
commit 62fe53b147
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
fastNLP/core/callback.py
View File

@ -464,7 +464,7 @@ class GradientClipCallback(Callback):
self.clip_fun = nn.utils.clip_grad_value_
else:
raise ValueError("Only supports `norm` or `value` right now.")
self.parameters = parameters
self.parameters = list(parameters)
self.clip_value = clip_value
def on_backward_end(self):