This model is based on the Chinese offline version of paraformator (model_id: speech_paraformater-large_asr_nat zh cn-16k common vocab8404 Pytoch). We export the model into three parts: 1. encoder.onnx; 2. predictor.onnx; 3. decoder.onnx. At the same time, we have added metadata information on encoder.onnx according to the format requirements of sherpa-onnx, so that the model can run without relying on configuration files. The relevant code can be found in add_metadata.py, use para_onnx_offline.py script can load test speech and this model for decoding testing. In addition, we also provide a quantized model for int8.