This repo contains MRC OSI code related to BERT quantization.

Reference NLP task is combined Intent-Slot classification, as in https://paperswithcode.com/paper/bert-for-joint-intent-classification-and-slot

References to interesting papers are collected in ./doc/QuantizationInNLP.md

To run the Docker environment shell, run:

$ ./rundocker.sh docker/murygin-bert.docker

and then cd murygin/bert. Pass --no-map-sockets in case of socket binding conflict.

To run the experiment, see ./murygin/bert/README.md.