未验证 提交 747df80d 编写于 作者: 武毅 提交者: GitHub

Merge pull request #3 from gongweibao/wuyi7539_3

Add results.
......@@ -48,16 +48,18 @@
### Different Pserver Number
- Trainer Count: 100
- Trainer Count: 60
- Batch Size: 128
- Metrics: mini-batch / sec
| PServer Count | 10 | 20 | 40 | 60 |
| PServer Count | 3 | 6 |10 | 20 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - |
| PaddlePaddle Fluid(should fix in next PR) | 589.1 | 592.6 | 656.4 | 655.8 |
| PaddlePaddle v2 | 593.4 | 791.3 | 729.7 | 821.7 |
| TensorFlow | - | - | - | - |
*The performance gap between Fuild and v2 comes from the network interference.*
## Steps to run the performance test
......
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