Hugging Face's logo

qualcomm
/
OpusMT-En-Zh

Text Generation
PyTorch
foundation
android
Model card Files
xet
 Community

OpusMT-En-Zh: Optimized for Qualcomm Devices

OpusMT English to Chinese translation model is a state-of-the-art neural machine translation system designed for translating English text into Chinese. This model is based on the Marian transformer architecture and has been optimized for edge inference by splitting into encoder and decoder components with modified attention mechanisms. It exhibits robust performance for real-world translation tasks, making it highly reliable for practical applications. The model supports input sequences up to 256 tokens and can generate Chinese translations with high accuracy.

This is based on the implementation of OpusMT-En-Zh found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime Precision Chipset SDK Versions Download
ONNX float Universal QAIRT 2.42, ONNX Runtime 1.24.1 Download
QNN_CONTEXT_BINARY float qualcomm-qcs8275 QAIRT 2.42 Download
QNN_DLC float Universal QAIRT 2.43 Download
TFLITE float Universal QAIRT 2.43, TFLite 2.17.0 Download

For more device-specific assets and performance metrics, visit OpusMT-En-Zh on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

  • Custom weights (e.g., fine-tuned checkpoints)
  • Custom input shapes
  • Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for OpusMT-En-Zh on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

  • Model checkpoint: Helsinki-NLP/opus-mt-en-zh
  • Input resolution: 256 tokens (English text)
  • Max input sequence length: 256 tokens
  • Max output sequence length: 256 tokens
  • Number of parameters (OpusMTEncoder): ~74M
  • Model size (OpusMTEncoder) (float): ~280 MB
  • Number of parameters (OpusMTDecoder): ~74M
  • Model size (OpusMTDecoder) (float): ~280 MB
  • Number of encoder layers: 6
  • Number of decoder layers: 6
  • Attention heads: 8
  • Hidden dimension: 512

Performance Summary

Model Runtime Precision Chipset Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit
OpusMTDecoder ONNX float Snapdragon® X2 Elite 1.737 ms 161 - 161 MB NPU
OpusMTDecoder ONNX float Snapdragon® X Elite 3.2 ms 160 - 160 MB NPU
OpusMTDecoder ONNX float Snapdragon® 8 Gen 3 Mobile 2.775 ms 0 - 399 MB NPU
OpusMTDecoder ONNX float Qualcomm® QCS8550 (Proxy) 3.94 ms 12 - 14 MB NPU
OpusMTDecoder ONNX float Qualcomm® QCS9075 4.239 ms 12 - 27 MB NPU
OpusMTDecoder ONNX float Snapdragon® 8 Elite For Galaxy Mobile 2.349 ms 0 - 429 MB NPU
OpusMTDecoder ONNX float Snapdragon® 8 Elite Gen 5 Mobile 2.233 ms 1 - 378 MB NPU
OpusMTDecoder QNN_DLC float Snapdragon® X2 Elite 2.071 ms 6 - 6 MB NPU
OpusMTDecoder QNN_DLC float Snapdragon® X Elite 2.965 ms 6 - 6 MB NPU
OpusMTDecoder QNN_DLC float Snapdragon® 8 Gen 3 Mobile 2.353 ms 6 - 193 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® QCS8275 (Proxy) 6.844 ms 2 - 167 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® QCS8550 (Proxy) 3.04 ms 1 - 2 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® SA8775P 3.956 ms 2 - 168 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® QCS9075 3.367 ms 8 - 16 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® QCS8450 (Proxy) 3.986 ms 4 - 173 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® SA7255P 6.844 ms 2 - 167 MB NPU
OpusMTDecoder QNN_DLC float Qualcomm® SA8295P 4.138 ms 6 - 157 MB NPU
OpusMTDecoder QNN_DLC float Snapdragon® 8 Elite For Galaxy Mobile 2.13 ms 0 - 167 MB NPU
OpusMTDecoder QNN_DLC float Snapdragon® 8 Elite Gen 5 Mobile 1.909 ms 0 - 171 MB NPU
OpusMTDecoder TFLITE float Snapdragon® 8 Gen 3 Mobile 2.773 ms 0 - 377 MB NPU
OpusMTDecoder TFLITE float Qualcomm® QCS8275 (Proxy) 6.721 ms 0 - 230 MB NPU
OpusMTDecoder TFLITE float Qualcomm® QCS8550 (Proxy) 3.424 ms 0 - 3 MB NPU
OpusMTDecoder TFLITE float Qualcomm® SA8775P 4.321 ms 0 - 229 MB NPU
OpusMTDecoder TFLITE float Qualcomm® QCS9075 4.205 ms 0 - 186 MB NPU
OpusMTDecoder TFLITE float Qualcomm® QCS8450 (Proxy) 4.709 ms 0 - 354 MB NPU
OpusMTDecoder TFLITE float Qualcomm® SA7255P 6.721 ms 0 - 230 MB NPU
OpusMTDecoder TFLITE float Qualcomm® SA8295P 4.918 ms 0 - 212 MB NPU
OpusMTDecoder TFLITE float Snapdragon® 8 Elite For Galaxy Mobile 2.52 ms 0 - 406 MB NPU
OpusMTDecoder TFLITE float Snapdragon® 8 Elite Gen 5 Mobile 2.132 ms 0 - 374 MB NPU
OpusMTEncoder ONNX float Snapdragon® X2 Elite 2.22 ms 108 - 108 MB NPU
OpusMTEncoder ONNX float Snapdragon® X Elite 5.061 ms 108 - 108 MB NPU
OpusMTEncoder ONNX float Snapdragon® 8 Gen 3 Mobile 3.397 ms 0 - 266 MB NPU
OpusMTEncoder ONNX float Qualcomm® QCS8550 (Proxy) 4.755 ms 0 - 123 MB NPU
OpusMTEncoder ONNX float Qualcomm® QCS9075 6.263 ms 16 - 18 MB NPU
OpusMTEncoder ONNX float Snapdragon® 8 Elite For Galaxy Mobile 2.777 ms 0 - 290 MB NPU
OpusMTEncoder ONNX float Snapdragon® 8 Elite Gen 5 Mobile 2.032 ms 0 - 257 MB NPU
OpusMTEncoder QNN_DLC float Snapdragon® X2 Elite 1.973 ms 0 - 0 MB NPU
OpusMTEncoder QNN_DLC float Snapdragon® X Elite 3.889 ms 0 - 0 MB NPU
OpusMTEncoder QNN_DLC float Snapdragon® 8 Gen 3 Mobile 2.542 ms 0 - 89 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® QCS8275 (Proxy) 12.607 ms 0 - 52 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® QCS8550 (Proxy) 3.521 ms 0 - 2 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® SA8775P 19.246 ms 0 - 52 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® QCS9075 4.675 ms 0 - 5 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® QCS8450 (Proxy) 4.768 ms 0 - 86 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® SA7255P 12.607 ms 0 - 52 MB NPU
OpusMTEncoder QNN_DLC float Qualcomm® SA8295P 5.349 ms 0 - 54 MB NPU
OpusMTEncoder QNN_DLC float Snapdragon® 8 Elite For Galaxy Mobile 2.139 ms 0 - 53 MB NPU
OpusMTEncoder QNN_DLC float Snapdragon® 8 Elite Gen 5 Mobile 1.612 ms 0 - 58 MB NPU
OpusMTEncoder TFLITE float Snapdragon® 8 Gen 3 Mobile 2.663 ms 0 - 260 MB NPU
OpusMTEncoder TFLITE float Qualcomm® QCS8275 (Proxy) 12.774 ms 6 - 99 MB NPU
OpusMTEncoder TFLITE float Qualcomm® QCS8550 (Proxy) 3.682 ms 0 - 3 MB NPU
OpusMTEncoder TFLITE float Qualcomm® SA8775P 4.836 ms 6 - 100 MB NPU
OpusMTEncoder TFLITE float Qualcomm® QCS9075 5.137 ms 6 - 120 MB NPU
OpusMTEncoder TFLITE float Qualcomm® QCS8450 (Proxy) 5.08 ms 6 - 258 MB NPU
OpusMTEncoder TFLITE float Qualcomm® SA7255P 12.774 ms 6 - 99 MB NPU
OpusMTEncoder TFLITE float Qualcomm® SA8295P 5.662 ms 6 - 95 MB NPU
OpusMTEncoder TFLITE float Snapdragon® 8 Elite For Galaxy Mobile 2.237 ms 0 - 279 MB NPU
OpusMTEncoder TFLITE float Snapdragon® 8 Elite Gen 5 Mobile 1.774 ms 0 - 243 MB NPU

License

  • The license for the original implementation of OpusMT-En-Zh can be found here.

References

Community

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
Text Generation
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support