valar000

Fix two error on pytorch0.4 cuda

3

Fix TypeError: multinomial() missing 1 required positional arguments:"num_samples" Fix RuntimeError: cudnn RNN backward can only be called in training mode

内存泄漏?

2

**Describe the bug** 启动一天之后 swap使用超过了1G **To Reproduce** Steps to reproduce the behavior: 1. 使用docker启动 2. 放置1天 **Expected behavior** 内存使用增长不大 **Screenshots**  **Desktop (please complete the following information):** - docker 启动...