echoht

can this code run on a cpu machine？

2

I run on a cpu machine with default args, I got the following errors： num_params: 167301123 INFO:root:Variables info has been saved. 5%|#####5 | 1533/30681 [3:21:15

why i have codecs problem in open(f) line? I use python 3.6

2

python version 3.6.0 the error line code is : _prepro = lambda x: [line.strip() for line in open(x, 'r').read().split("\n") \ if not line.startswith("

https://github.com/lajanugen/zeshel/blob/584f15d5b0ae285df7dcd1a3e99ffae2203984dc/run_classifier.py#L203 BertModel has no such param.

loss的代码关于batch size的处理有bug。

4

 这里batch size设置为1时，逻辑没有问题。当batch size！=1时，会出现tensor size不match的情况。

fix batch size bug

3

fix bug when batchsize is != 1.

我看了webRL和AutoGLM的Paper，autoGLM里提到使用webRL的训练框架，我猜它的数据集应该不是VAB开源的这部分数据集，极可能是涉及更多网站、特别是中文网站的任务吧？这部分数据有开源吗？

### Reminder - [x] I have read the above rules and searched the existing issues. ### System Info 引用代码处： ` # Run PPO step self.model.train() stats = self.step(queries, responses, rewards)...