VIT: Optimized for Qualcomm Devices
VIT is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.
This is based on the implementation of VIT 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 1.27.1 | Download |
| ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | Download |
| ONNX | w8a8_mixed_int16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
| TFLITE | w8a8 | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit VIT 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 VIT on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.image_classification
Model Stats:
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 86.6M
- Model size (float): 330 MB
- Model size (w8a16): 86.2 MB
- Model size (w8a8): 83.2 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| VIT | ONNX | float | Snapdragon® X2 Elite | 3.051 ms | 2 - 2 MB | NPU |
| VIT | ONNX | float | Snapdragon® X Elite | 7.513 ms | 170 - 170 MB | NPU |
| VIT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.013 ms | 0 - 304 MB | NPU |
| VIT | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 12.197 ms | 1 - 283 MB | NPU |
| VIT | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.169 ms | 0 - 320 MB | NPU |
| VIT | ONNX | float | Qualcomm® QCS8450 | 12.197 ms | 1 - 283 MB | NPU |
| VIT | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 9.54 ms | 0 - 4 MB | NPU |
| VIT | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 7.513 ms | 170 - 170 MB | NPU |
| VIT | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 3.462 ms | 0 - 178 MB | NPU |
| VIT | ONNX | float | Snapdragon® 8 Elite Mobile | 3.462 ms | 0 - 178 MB | NPU |
| VIT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.811 ms | 0 - 161 MB | NPU |
| VIT | ONNX | w8a16 | Snapdragon® X2 Elite | 2.01 ms | 1 - 1 MB | NPU |
| VIT | ONNX | w8a16 | Snapdragon® X Elite | 4.759 ms | 86 - 86 MB | NPU |
| VIT | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.112 ms | 0 - 314 MB | NPU |
| VIT | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 6.433 ms | 0 - 316 MB | NPU |
| VIT | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 21.374 ms | 0 - 3 MB | NPU |
| VIT | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.568 ms | 0 - 3 MB | NPU |
| VIT | ONNX | w8a16 | Qualcomm® QCS8450 | 6.433 ms | 0 - 316 MB | NPU |
| VIT | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 4.451 ms | 0 - 3 MB | NPU |
| VIT | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 4.759 ms | 86 - 86 MB | NPU |
| VIT | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.444 ms | 0 - 261 MB | NPU |
| VIT | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 2.444 ms | 0 - 261 MB | NPU |
| VIT | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 1.794 ms | 0 - 269 MB | NPU |
| VIT | ONNX | w8a8 | Snapdragon® X2 Elite | 4.123 ms | 1 - 1 MB | NPU |
| VIT | ONNX | w8a8 | Snapdragon® X Elite | 9.747 ms | 85 - 85 MB | NPU |
| VIT | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.285 ms | 0 - 335 MB | NPU |
| VIT | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 13.151 ms | 0 - 340 MB | NPU |
| VIT | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 62.502 ms | 0 - 3 MB | NPU |
| VIT | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 9.377 ms | 0 - 106 MB | NPU |
| VIT | ONNX | w8a8 | Qualcomm® QCS8450 | 13.151 ms | 0 - 340 MB | NPU |
| VIT | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 8.112 ms | 0 - 3 MB | NPU |
| VIT | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 9.747 ms | 85 - 85 MB | NPU |
| VIT | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 5.407 ms | 0 - 239 MB | NPU |
| VIT | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 5.407 ms | 0 - 239 MB | NPU |
| VIT | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.515 ms | 0 - 246 MB | NPU |
| VIT | ONNX | w8a8_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 57.294 ms | 123 - 563 MB | NPU |
| VIT | ONNX | w8a8_mixed_int16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 70.971 ms | 118 - 121 MB | NPU |
| VIT | ONNX | w8a8_mixed_int16 | Qualcomm® Dragonwing™ IQ-9075 | 84.027 ms | 125 - 131 MB | NPU |
| VIT | ONNX | w8a8_mixed_int16 | Qualcomm® Dragonwing™ Q-8750 | 48.304 ms | 127 - 465 MB | NPU |
| VIT | ONNX | w8a8_mixed_int16 | Snapdragon® 8 Elite Mobile | 48.304 ms | 127 - 465 MB | NPU |
| VIT | ONNX | w8a8_mixed_int16 | Snapdragon® 8 Elite Gen 5 Mobile | 38.152 ms | 123 - 476 MB | NPU |
| VIT | QNN_DLC | float | Snapdragon® X2 Elite | 3.493 ms | 1 - 1 MB | NPU |
| VIT | QNN_DLC | float | Snapdragon® X Elite | 8.358 ms | 1 - 1 MB | NPU |
| VIT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 5.403 ms | 0 - 253 MB | NPU |
| VIT | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 13.034 ms | 1 - 229 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 34.784 ms | 1 - 168 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.703 ms | 1 - 400 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® SA8775P | 10.471 ms | 1 - 169 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® SA8650P | 10.471 ms | 1 - 169 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® SA8255P | 10.471 ms | 1 - 169 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® QCS8450 | 13.034 ms | 1 - 229 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 10.169 ms | 3 - 5 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 8.358 ms | 1 - 1 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 3.751 ms | 1 - 157 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® SA7255P | 34.784 ms | 1 - 168 MB | NPU |
| VIT | QNN_DLC | float | Qualcomm® SA8295P | 12.942 ms | 1 - 145 MB | NPU |
| VIT | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 3.751 ms | 1 - 157 MB | NPU |
| VIT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.946 ms | 1 - 156 MB | NPU |
| VIT | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 3.596 ms | 0 - 0 MB | NPU |
| VIT | QNN_DLC | w8a16 | Snapdragon® X Elite | 8.219 ms | 0 - 0 MB | NPU |
| VIT | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 5.088 ms | 0 - 319 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 17.354 ms | 0 - 258 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.6 ms | 0 - 2 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® SA8775P | 7.778 ms | 0 - 259 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® SA8650P | 7.778 ms | 0 - 259 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® SA8255P | 7.778 ms | 0 - 259 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 7.607 ms | 0 - 2 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 8.219 ms | 0 - 0 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 117.255 ms | 1 - 405 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 12.881 ms | 0 - 403 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 3.904 ms | 0 - 260 MB | NPU |
| VIT | QNN_DLC | w8a16 | Qualcomm® SA7255P | 17.354 ms | 0 - 258 MB | NPU |
| VIT | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 3.904 ms | 0 - 260 MB | NPU |
| VIT | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.867 ms | 0 - 279 MB | NPU |
| VIT | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 12.881 ms | 0 - 403 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 4.448 ms | 0 - 0 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® X Elite | 10.216 ms | 0 - 0 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.482 ms | 0 - 305 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 13.207 ms | 0 - 314 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 62.275 ms | 2 - 4 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8275 | 28.467 ms | 0 - 201 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 9.595 ms | 0 - 2 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® SA8775P | 8.67 ms | 0 - 203 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® SA8650P | 8.67 ms | 0 - 203 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® SA8255P | 8.67 ms | 0 - 203 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 13.207 ms | 0 - 314 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 8.366 ms | 0 - 2 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 10.216 ms | 0 - 0 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 212.341 ms | 0 - 496 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 19.182 ms | 0 - 358 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 5.385 ms | 0 - 203 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® SA7255P | 28.467 ms | 0 - 201 MB | NPU |
| VIT | QNN_DLC | w8a8 | Qualcomm® SA8295P | 15.601 ms | 0 - 211 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 5.385 ms | 0 - 203 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.546 ms | 0 - 213 MB | NPU |
| VIT | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 19.182 ms | 0 - 358 MB | NPU |
| VIT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.284 ms | 0 - 309 MB | NPU |
| VIT | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 12.955 ms | 0 - 288 MB | NPU |
| VIT | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 34.607 ms | 0 - 168 MB | NPU |
| VIT | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.235 ms | 0 - 3 MB | NPU |
| VIT | TFLITE | float | Qualcomm® SA8775P | 10.164 ms | 0 - 186 MB | NPU |
| VIT | TFLITE | float | Qualcomm® SA8650P | 10.164 ms | 0 - 186 MB | NPU |
| VIT | TFLITE | float | Qualcomm® SA8255P | 10.164 ms | 0 - 186 MB | NPU |
| VIT | TFLITE | float | Qualcomm® QCS8450 | 12.955 ms | 0 - 288 MB | NPU |
| VIT | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 9.841 ms | 0 - 174 MB | NPU |
| VIT | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 3.643 ms | 0 - 177 MB | NPU |
| VIT | TFLITE | float | Qualcomm® SA7255P | 34.607 ms | 0 - 168 MB | NPU |
| VIT | TFLITE | float | Qualcomm® SA8295P | 13.268 ms | 0 - 156 MB | NPU |
| VIT | TFLITE | float | Snapdragon® 8 Elite Mobile | 3.643 ms | 0 - 177 MB | NPU |
| VIT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.031 ms | 0 - 171 MB | NPU |
| VIT | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 8.697 ms | 0 - 488 MB | NPU |
| VIT | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 20.112 ms | 0 - 447 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 139.29 ms | 1 - 100 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 12.364 ms | 0 - 3 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® SA8775P | 12.214 ms | 0 - 388 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® SA8650P | 12.214 ms | 0 - 388 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® SA8255P | 12.214 ms | 0 - 388 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® QCS8450 | 20.112 ms | 0 - 447 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 13.986 ms | 0 - 88 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 199.781 ms | 2 - 282 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 29.603 ms | 2 - 216 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 6.93 ms | 0 - 379 MB | NPU |
| VIT | TFLITE | w8a8 | Qualcomm® SA8295P | 19.272 ms | 0 - 346 MB | NPU |
| VIT | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 6.93 ms | 0 - 379 MB | NPU |
| VIT | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 4.602 ms | 0 - 382 MB | NPU |
| VIT | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 29.603 ms | 2 - 216 MB | NPU |
License
- The license for the original implementation of VIT can be found here.
References
- An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
- Source Model Implementation
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.
