Skip to content

Commit

Permalink
add e2e_time for GLM 1x1
Browse files Browse the repository at this point in the history
  • Loading branch information
zhouyu committed Aug 18, 2023
1 parent b38f10d commit 4c38f32
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion training/nvidia/glm-pytorch/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,6 @@
| ------------------- | --------- | --------------- | -------- | ------- | ------- | ------ | ----- | --------- | ----- |
| A100单机8卡(1x8) | fp32 | / | 2763 | 36.5 | 42.4 | 42.4 | 0.808 | 33.0/40.0 | 0.035 |
| A100单机8卡(1x8) | fp32 | bs=16, lr=1e-05 | 2688 | 37.4 | 43.5 | 43.5 | 0.801 | 39.5/40.0 | 0.035 |
| A100单机单卡(1x1) | fp32 | bs=16, lr=1e-05 | | 0.35 | 5.5 | 5.5 | | 35.0/40.0 | |
| A100单机单卡(1x1) | fp32 | bs=16, lr=1e-05 | 1169 | 0.35 | 5.5 | 5.5 | | 35.0/40.0 | 0.036 |

> 注:使用GLMForMultiTokenCloze进行forward计算你,得到MFU=0.04, 使用GLMModel模型forward计算,得到MFU=0.08. 本模型的MFU值偏低是由于原始模型的MFU较低。

0 comments on commit 4c38f32

Please sign in to comment.