Code for joint NLU model

This implementation follows the work of BERT for Joint Intent Classification and Slot Filling.

Usage

Use the script trian.py to launch the Single-GPU training:

python3 train.py

Use the script run_train.sh to launch the Multi-GPT training:

bash run_training.sh

Data Processing

The data from the SMP2019 task is provided in the folder data.

TODO

• Implement CRF layer. Code from this repo can be referred to add the CRF layer.
• 调整intent loss 和 slot loss的比例