search

KhalilMrini / LAL-Parser

Neural Adobe-UCSD Parser, the current State of the Art in Constituency and Dependency Parsing.
138 stars 24 forks source link

Slow parsing speed #7

Closed ZhaofengWu closed 4 years ago

ZhaofengWu commented 4 years ago

Hi, thanks for publishing the code and the model! I'm trying to use the model to parse some sentences using the parse.sh script, but it's been very slow. With the default eval batch size of 50, it is taking 100-400s per batch on an RTX 8000. Is that expected?

KhalilMrini commented 4 years ago

Hi Zhaofeng, thank you so much for your interest in our work! I am really pleased that you are using our model. Actually, the parse.sh script computes head contributions, which takes a lot of time. I added a script parse_quick.sh that does not compute head contributions, and just parses sentences. I hope that helps!