LeViT: Optimized for Qualcomm Devices
LeViT is a vision transformer model that can classify images from the Imagenet dataset.
This is based on the implementation of LeViT 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 2.45, ONNX Runtime 1.25.0 | Download |
| ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.25.0 | Download |
| ONNX | w8a16_mixed_int16 | Universal | QAIRT 2.45, ONNX Runtime 1.25.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a16_mixed_int16 | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit LeViT 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 LeViT on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.image_classification
Model Stats:
- Model checkpoint: LeViT-128S
- Input resolution: 224x224
- Number of parameters: 7.82M
- Model size (float): 29.9 MB
- Model size (w8a16): 8.83 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| LeViT | ONNX | float | Snapdragon® X2 Elite | 0.687 ms | 212 - 212 MB | NPU |
| LeViT | ONNX | float | Snapdragon® X Elite | 1.318 ms | 181 - 181 MB | NPU |
| LeViT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.866 ms | 1 - 74 MB | NPU |
| LeViT | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 1.589 ms | 1 - 74 MB | NPU |
| LeViT | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1.278 ms | 0 - 35 MB | NPU |
| LeViT | ONNX | float | Qualcomm® QCS8450 | 1.589 ms | 1 - 74 MB | NPU |
| LeViT | ONNX | float | Snapdragon® 8 Elite Mobile | 0.704 ms | 0 - 57 MB | NPU |
| LeViT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.651 ms | 1 - 52 MB | NPU |
| LeViT | ONNX | float | Qualcomm® QCS9075 | 1.62 ms | 1 - 46 MB | NPU |
| LeViT | ONNX | float | Qualcomm® QCS8750 | 0.704 ms | 0 - 57 MB | NPU |
| LeViT | ONNX | float | Qualcomm® QCS7181 | 1.318 ms | 181 - 181 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® X2 Elite | 0.486 ms | 212 - 212 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® X Elite | 1.057 ms | 149 - 149 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.696 ms | 0 - 77 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 1.258 ms | 0 - 75 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS6490 | 2.878 ms | 0 - 45 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 1.025 ms | 0 - 13 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS8450 | 1.258 ms | 0 - 75 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.474 ms | 0 - 53 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 0.52 ms | 0 - 54 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS9075 | 1.231 ms | 0 - 45 MB | NPU |
| LeViT | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1.013 ms | 0 - 52 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCM6690 | 4.323 ms | 0 - 175 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS7790 | 1.013 ms | 0 - 52 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS8750 | 0.52 ms | 0 - 54 MB | NPU |
| LeViT | ONNX | w8a16 | Qualcomm® QCS7181 | 1.057 ms | 149 - 149 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Snapdragon® X2 Elite | 0.507 ms | 213 - 213 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Snapdragon® X Elite | 1.12 ms | 149 - 149 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 0.732 ms | 0 - 77 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Qualcomm® QCS8550 (Proxy) | 1.054 ms | 0 - 18 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Qualcomm® QCM6690 | 4.775 ms | 0 - 179 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.498 ms | 0 - 59 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Qualcomm® QCS9075 | 1.313 ms | 0 - 45 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Elite Mobile | 0.545 ms | 0 - 59 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Snapdragon® 7 Gen 4 Mobile | 1.062 ms | 0 - 55 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Qualcomm® QCS7790 | 1.062 ms | 0 - 55 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Qualcomm® QCS8750 | 0.545 ms | 0 - 59 MB | NPU |
| LeViT | ONNX | w8a16_mixed_int16 | Qualcomm® QCS7181 | 1.12 ms | 149 - 149 MB | NPU |
| LeViT | QNN_DLC | float | Snapdragon® X2 Elite | 0.974 ms | 1 - 1 MB | NPU |
| LeViT | QNN_DLC | float | Snapdragon® X Elite | 1.835 ms | 1 - 1 MB | NPU |
| LeViT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.077 ms | 0 - 86 MB | NPU |
| LeViT | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 2.372 ms | 1 - 84 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® QCS8275 | 3.801 ms | 1 - 58 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.599 ms | 1 - 2 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® QCS8450 | 2.372 ms | 1 - 84 MB | NPU |
| LeViT | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.831 ms | 0 - 58 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® SA7255P | 3.801 ms | 1 - 58 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® SA8295P | 2.362 ms | 0 - 52 MB | NPU |
| LeViT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.722 ms | 1 - 62 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® QCS9075 | 1.882 ms | 3 - 5 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® QCS8750 | 0.831 ms | 0 - 58 MB | NPU |
| LeViT | QNN_DLC | float | Qualcomm® QCS7181 | 1.835 ms | 1 - 1 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.807 ms | 0 - 0 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Snapdragon® X Elite | 1.61 ms | 0 - 0 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.939 ms | 0 - 71 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCS8275 | 2.783 ms | 0 - 50 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 1.389 ms | 0 - 14 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.57 ms | 0 - 52 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 0.663 ms | 0 - 47 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 1.716 ms | 0 - 2 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1.416 ms | 0 - 50 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 5.192 ms | 0 - 174 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® SA7255P | 2.783 ms | 0 - 50 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCS7790 | 1.416 ms | 0 - 50 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCS8750 | 0.663 ms | 0 - 47 MB | NPU |
| LeViT | QNN_DLC | w8a16 | Qualcomm® QCS7181 | 1.61 ms | 0 - 0 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Snapdragon® X2 Elite | 0.812 ms | 0 - 0 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Snapdragon® X Elite | 1.654 ms | 0 - 0 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 0.979 ms | 0 - 72 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS8275 | 2.896 ms | 0 - 51 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS8550 (Proxy) | 1.468 ms | 0 - 30 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCM6690 | 5.957 ms | 0 - 178 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.613 ms | 0 - 53 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS9075 | 1.7 ms | 0 - 2 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Elite Mobile | 0.698 ms | 0 - 54 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 7 Gen 4 Mobile | 1.436 ms | 0 - 51 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® SA7255P | 2.896 ms | 0 - 51 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS7790 | 1.436 ms | 0 - 51 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS8750 | 0.698 ms | 0 - 54 MB | NPU |
| LeViT | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS7181 | 1.654 ms | 0 - 0 MB | NPU |
| LeViT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.027 ms | 19 - 111 MB | NPU |
| LeViT | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 2.361 ms | 0 - 84 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® QCS8275 | 4.011 ms | 0 - 62 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.514 ms | 0 - 3 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® SA8775P | 7.832 ms | 1 - 7 MB | CPU |
| LeViT | TFLITE | float | Qualcomm® SA8650P | 7.832 ms | 1 - 7 MB | CPU |
| LeViT | TFLITE | float | Qualcomm® SA8255P | 7.832 ms | 1 - 7 MB | CPU |
| LeViT | TFLITE | float | Qualcomm® QCS8450 | 2.361 ms | 0 - 84 MB | NPU |
| LeViT | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.783 ms | 0 - 67 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® SA7255P | 4.011 ms | 0 - 62 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® SA8295P | 2.36 ms | 0 - 53 MB | NPU |
| LeViT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.661 ms | 0 - 61 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® QCS9075 | 1.857 ms | 0 - 19 MB | NPU |
| LeViT | TFLITE | float | Qualcomm® QCS8750 | 0.783 ms | 0 - 67 MB | NPU |
License
- The license for the original implementation of LeViT can be found here.
References
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.
