google-research / bert

TensorFlow code and pre-trained models for BERT
https://arxiv.org/abs/1810.04805
Apache License 2.0
38.12k stars 9.6k forks source link

Speedup using Transformer-XL #356

Open kaushalshetty opened 5 years ago

kaushalshetty commented 5 years ago

Prediction in CPU takes about 1 sec for a paragraph. Would using transformer-xl speedup predictions on CPU?

zihaolucky commented 5 years ago

Also interest in this topic.

xwzhong commented 5 years ago

also interest

gallupliu commented 5 years ago

also interest

artemisart commented 5 years ago

Transformer-XL is not bidirectional and the speed-up mainly makes sense for next word prediction.

ArthurRizar commented 5 years ago

Also interest

Single430 commented 5 years ago

also interest