Yang Liu

Results 11 comments of Yang Liu

请问解决了么?修改超参数的地方在对应的yaml文件中

训练集指的那两个合成数据集吗,四张a100需要一天多

那个是作为测试集吧,训练还是合成的

这里都是测试集,训练集都是syth和mj,论文中有提到

在这打断点后, ![image](https://user-images.githubusercontent.com/69834766/201532607-347419ef-df78-4370-802e-bc9df1c47951.png) 两种方法print出来的数值不一样: ![image](https://user-images.githubusercontent.com/69834766/201532635-8aaeef06-b6b4-44ef-9db5-e9fa5f4bbd1c.png)

未修改过任何代码,怀疑跟paddle和驱动/显卡/docker的兼容适配有关

把 use_gpu 关了是可以正常训练的

换了[2.1.3-gpu-cuda11.2-cudnn8](https://hub.docker.com/layers/paddlepaddle/paddle/2.1.3-gpu-cuda11.2-cudnn8/images/sha256-ad3b494a60a52390199902e284c1d87e40defe390ca702619bde51f5aa904364?context=explore), 问题解决.

I also have the same need as you, have you solved it now?

I am trying to follow this work recently. Can you share your training code?