BERT Base NER (IGIE)

Model Description

BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications.

Supported Environments

Model Preparation

Prepare Resources

Pretrained model: https://huggingface.co/bert-base-chinese

Dataset: http://s3.bmio.net/kashgari/china-people-daily-ner-corpus.tar.gz

Install Dependencies

pip3 install -r requirements.txt

Model Conversion

export DATASETS_DIR=/Path/to/china-people-daily-ner-corpus/

# Get pytorch weights
python3 get_weights.py

# Do QAT for INT8 test, will take a long time  
cd Int8QAT/
python3 run_qat.py --model_dir ../test/ --datasets_dir ${DATASETS_DIR}
python3 export_hdf5.py --model quant_base/pytorch_model.bin
cd ../

Model Inference

INT8

# Accuracy
bash scripts/infer_bert_base_ner_int8_accuracy.sh
# Performance
bash scripts/infer_bert_base_ner_int8_performance.sh

Model Results