提交 fb26ca60 编写于 作者: Z zhangxuefei

Merge branch 'release/v1.2' into develop

......@@ -12,5 +12,5 @@ pandas
flask >= 1.1.0
flake8
tb-paddle
tb-nightly
tensorboard >= 1.15
cma == 2.7.0
......@@ -11,5 +11,5 @@ pandas < 0.25.0
flake8
flask >= 1.1.0
tb-paddle
tb-nightly
tensorboard >= 1.15
cma == 2.7.0
......@@ -32,7 +32,7 @@ max_version, mid_version, min_version = python_version()
REQUIRED_PACKAGES = [
'six >= 1.10.0', 'protobuf >= 3.6.0', 'pyyaml', 'Pillow', 'requests',
'tb-paddle', 'tb-nightly', 'cma == 2.7.0', 'flask >= 1.1.0'
'tb-paddle', 'tensorboard >= 1.15', 'cma == 2.7.0', 'flask >= 1.1.0'
]
if max_version < 3:
......
......@@ -188,7 +188,7 @@ PaddleHub中集成了ERNIE、BERT、LAC、ELMo等[NLP预训练模型](https://ww
slanted triangular learning rate和warm up + linear decay在原理上和实验结果上都是相似的,Discriminative fine-tuning和Gradual unfreezing微调策略在使用中,应当注意它们会降低模型的拟合能力,可以适当提高训练的轮数。
PaddleHub 1.2已发布AutoDL Finetuner,可以自动搜索超参设置,详情请参考[PaddleHub AutoDL Finetuner](./autofinetune.md)。如有任何疑问欢迎您在issues中向我们提出!
PaddleHub 1.2已发布AutoDL Finetuner,可以自动搜索超参设置,详情请参考[PaddleHub AutoDL Finetuner](https://github.com/PaddlePaddle/PaddleHub/blob/develop/tutorial/autofinetune.md)。如有任何疑问欢迎您在issues中向我们提出!
## 六、参考文献
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册