deeplearning4j / deeplearning4j/deeplearning4j-examples
tf1.4(pb) embedding operator is slow
Open
- Dominant language
- Java
- Stars
- 2.5k
- Forks
- 1.8k
- PR merge metrics
- No merged PRs in 30d
Description
Hi，
If I use tf1.4 model with pb format, it takes almost 200ms in DeepFM (70 features, 3 dim each embedding lookup vector). After debug, we catch the problems in embedding operator
Contributor guide
Assessment
This issue has not been assessed yet.