Skip to content

Conversation

@wozna
Copy link
Contributor

@wozna wozna commented Feb 23, 2021

PR types

Others

PR changes

Others

Describe

This PR adds the bfloat16 GRU model test.
The results for the BF16 on the CooperLake are as follows:

batch size = 50, iterations = 160

Native FP32 FP32 BF16 (BF16/FP32)
FPS (1 thread) 2794.97 2700.45 4210.27 1.56
FPS (4 threads) 3076.66 4756.45 6186.94 1.30
GRU model FP32 BF16 diff
Precision 0.89211 0.89225 -0.00014
Recall 0.89442 0.89457 -0.00015
F1 score 0.89326 0.89341 -0.00015

Copy link
Contributor

@jczaja jczaja left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@wozna
Copy link
Contributor Author

wozna commented Feb 24, 2021

@luotao1 Could you please start your review?

@luotao1 luotao1 merged commit d116024 into PaddlePaddle:develop Feb 25, 2021
@wozna wozna deleted the bf16_gru branch February 24, 2023 16:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants