anchor-flux/punct-ct-transformer-zh

Introduction

This model is converted from https://modelscope.cn/models/iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch/summary

The commands to generate the onnx model are given below:

pip install funasr modelscope
pip install kaldi-native-fbank torchaudio onnx onnxruntime

mkdir -p /tmp/models
cd /tmp/models

git clone https://www.modelscope.cn/iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch.git
cd punc_ct-transformer_zh-cn-common-vocab272727-pytorch
git lfs pull --include model.pt

cd /tmp
git clone https://github.com/alibaba-damo-academy/FunASR
cd FunASR/runtime/python/onnxruntime

cat >export-onnx.py <<EOF

from funasr_onnx import CT_Transformer
model_dir = "/tmp/punc_ct-transformer_zh-cn-common-vocab272727-pytorch" # model = CT_Transformer(model_dir, quantize=True) model = CT_Transformer(model_dir)
EOF

chmod +x export-onnx.py

./export-onnx.py

You will find the exported model.onnx file inside /tmp/models/punc_ct-transformer_zh-cn-common-vocab272727-pytorch.

Now you can use ./add-model-metadata.py in this repo to add metadata to the generated model.onnx.

You can use

from onnxruntime.quantization import QuantType, quantize_dynamic

quantize_dynamic(
  model_input="./model.onnx",
  model_output="./model.int8.onnx",
  weight_type=QuantType.QUInt8,
)

to get the int8 quantized model

-rw-r--r--  1 fangjun  staff    72M Jun 18 10:33 model.int8.onnx
-rw-r--r--  1 fangjun  staff   281M Apr 12  2024 model.onnx