search

kentonl / e2e-coref

End-to-end Neural Coreference Resolution
Apache License 2.0
524 stars 173 forks source link

Memory Usage Inquiries #1

Closed theSage21 closed 7 years ago

theSage21 commented 7 years ago

First off thank you for the code! I wanted to know your thoughts about bringing down the memory usage while running demo.py. Am I right in assuming that not using the full glove set will bring down the performance since?

kentonl commented 7 years ago

Yes, the full glove embeddings are a hefty chunk of the memory usage. To verify this, you can try truncating the embeddings file to the first 30k lines or so. Memory usage should drop significantly while accuracy should still be decent.

theSage21 commented 7 years ago

Yes. I've tried that. With 30k memory use is around 2.4 GiB

Thanks a lot for your work once again!