Modelpublic

ranger810/sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12-int8

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes43downloads

Model Card

sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12-int8

A Chinese/English punctuation restoration model based on the Alibaba DAMO Academy CT-Transformer architecture, exported to ONNX INT8 quantized format using the sherpa-onnx framework.

Model Overview

Attribute	Details
Original Model	iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch
Model Format	ONNX INT8 Quantized
Supported Languages	Chinese, English
Vocabulary Size	272,727
Framework	sherpa-onnx
License	Apache 2.0

File List

Filename	Description
`model.int8.onnx`	INT8 quantized ONNX model file
`tokens.json`	Vocabulary file
`config.yaml`	Model configuration file
`test.py`	Test script
`show-model-input-output.py`	Script to inspect model input/output structure
`add-model-metadata.py`	Script to add model metadata

Usage

Install sherpa-onnx

bash

pip install sherpa-onnx

Python Example

python

import sherpa_onnx

config = sherpa_onnx.OfflinePunctuationConfig(
    model=sherpa_onnx.OfflinePunctuationModelConfig(
        ct_transformer="./model.int8.onnx",
    ),
    tokens="./tokens.json",
)

punct = sherpa_onnx.OfflinePunctuation(config)

text = "我们都是中国人我爱中国"
result = punct.add_punctuation(text)
print(result)
# Output: 我们都是中国人，我爱中国。

Run Test Script

bash

python test.py

Model Origin

This model is converted from the following original model by the sherpa-onnx project:

—Original Model: ModelScope - iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch
—Conversion: Exported via sherpa-onnx export scripts with INT8 quantization
—Original Paper: CT-Transformer: Controllable Time-delay Transformer for Real-Time Punctuation Prediction and Disfluency Detection

License

This model is licensed under the Apache License 2.0.

The original model is copyright of Alibaba DAMO Academy and is also released under the Apache 2.0 License.