thunlp / Few-NERD

Code and data of ACL 2021 paper "Few-NERD: A Few-shot Named Entity Recognition Dataset"
https://ningding97.github.io/fewnerd
Apache License 2.0
386 stars 54 forks source link

High CPU No CPU usage -> Sampler while loop problem #9

Closed laiviet closed 3 years ago

laiviet commented 3 years ago

Thanks for your realease.

While I try to run the code, it loads the model to the GPU (approx 2GB for XLMR) then it run mostly on CPU (20 processes at 100%). Is this a thing when it try to decode using Viterbi.

Thanks!

laiviet commented 3 years ago

This seems to be a problem in the Sampler when the number of samples in a class is too small that the while loop run infinitely. PRoblem solved.