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 2.50, ONNX Runtime 1.30.0	Download
ONNX	w8a16	Universal	QAIRT 2.50, ONNX Runtime 1.30.0	Download
ONNX	w8a8	Universal	QAIRT 2.50, ONNX Runtime 1.30.0	Download
ONNX	w8a8_mixed_int16	Universal	QAIRT 2.50, ONNX Runtime 1.30.0	Download
QNN_DLC	float	Universal	QAIRT 2.50	Download
QNN_DLC	w8a16	Universal	QAIRT 2.50	Download
QNN_DLC	w8a8	Universal	QAIRT 2.50	Download
QNN_DLC	w8a8_mixed_int16	Universal	QAIRT 2.50	Download
TFLITE	float	Universal	QAIRT 2.50	Download
TFLITE	w8a8	Universal	QAIRT 2.50	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:

Input resolution: 224x224
Model checkpoint: Imagenet
Model size (float): 330 MB
Model size (w8a16): 86.2 MB
Model size (w8a8): 83.2 MB
Number of parameters: 86.6M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
VIT	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	3.428 ms	0 - 158 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	4.417 ms	0 - 159 MB	NPU
VIT	ONNX	float	Snapdragon® X2 Elite	3.756 ms	2 - 2 MB	NPU
VIT	ONNX	float	Snapdragon® X Elite	9.409 ms	171 - 171 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	6.329 ms	0 - 287 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Gen 1 Mobile	13.405 ms	1 - 283 MB	NPU
VIT	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	12.426 ms	0 - 5 MB	NPU
VIT	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	8.976 ms	0 - 194 MB	NPU
VIT	ONNX	float	Qualcomm® QCS8450	13.405 ms	1 - 283 MB	NPU
VIT	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	11.876 ms	0 - 4 MB	NPU
VIT	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	9.409 ms	171 - 171 MB	NPU
VIT	ONNX	float	Qualcomm® Dragonwing™ Q-8750	4.417 ms	0 - 159 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	2.309 ms	0 - 194 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	3.161 ms	0 - 190 MB	NPU
VIT	ONNX	w8a16	Snapdragon® X2 Elite	2.551 ms	1 - 1 MB	NPU
VIT	ONNX	w8a16	Snapdragon® X Elite	6.174 ms	86 - 86 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	4.062 ms	0 - 289 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Gen 1 Mobile	7.816 ms	0 - 289 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ QCS6490	29.753 ms	0 - 4 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-8275	5.543 ms	0 - 5 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	5.927 ms	0 - 98 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCS8450	7.816 ms	0 - 289 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	5.856 ms	0 - 4 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	6.174 ms	86 - 86 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ Q-6690	98.115 ms	0 - 441 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	12.333 ms	0 - 405 MB	NPU
VIT	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	3.161 ms	0 - 190 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	12.333 ms	0 - 405 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	4.015 ms	0 - 184 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	5.81 ms	0 - 179 MB	NPU
VIT	ONNX	w8a8	Snapdragon® X2 Elite	4.604 ms	1 - 1 MB	NPU
VIT	ONNX	w8a8	Snapdragon® X Elite	10.805 ms	86 - 86 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	7.027 ms	0 - 298 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	13.964 ms	0 - 308 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ QCS6490	56.25 ms	0 - 3 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-8275	9.85 ms	0 - 4 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.327 ms	0 - 98 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCS8450	13.964 ms	0 - 308 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-9075	9.362 ms	0 - 3 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-X7181	10.805 ms	86 - 86 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ Q-6690	182.963 ms	0 - 509 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ Q-7790	20.153 ms	0 - 308 MB	NPU
VIT	ONNX	w8a8	Qualcomm® Dragonwing™ Q-8750	5.81 ms	0 - 179 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	20.153 ms	0 - 308 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	10.925 ms	47 - 374 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	13.469 ms	45 - 357 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® X2 Elite	12.546 ms	71 - 71 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® X Elite	24.689 ms	87 - 87 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Gen 3 Mobile	16.916 ms	47 - 480 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Gen 1 Mobile	28.587 ms	41 - 475 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® Dragonwing™ IQ-8275	26.884 ms	43 - 50 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	22.409 ms	43 - 46 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS8450	28.587 ms	41 - 475 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® Dragonwing™ IQ-9075	25.287 ms	48 - 55 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® Dragonwing™ IQ-X7181	24.689 ms	87 - 87 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® Dragonwing™ Q-8750	13.469 ms	45 - 357 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	5.14 ms	1 - 159 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	6.111 ms	1 - 158 MB	NPU
VIT	QNN_DLC	float	Snapdragon® X2 Elite	5.65 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Snapdragon® X Elite	12.614 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.2 ms	0 - 289 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	14.866 ms	1 - 290 MB	NPU
VIT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	14.777 ms	1 - 4 MB	NPU
VIT	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.764 ms	1 - 2 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8650P	14.585 ms	1 - 157 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8255P	14.585 ms	1 - 157 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8450	14.866 ms	1 - 290 MB	NPU
VIT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	14.877 ms	1 - 3 MB	NPU
VIT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	12.614 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	6.111 ms	1 - 158 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA7255P	42.138 ms	1 - 157 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8295P	16.035 ms	1 - 154 MB	NPU
VIT	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	2.873 ms	0 - 181 MB	NPU
VIT	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	3.925 ms	0 - 178 MB	NPU
VIT	QNN_DLC	w8a16	Snapdragon® X2 Elite	3.478 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a16	Snapdragon® X Elite	8.189 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	5.217 ms	0 - 280 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-8275	6.989 ms	0 - 3 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	7.628 ms	0 - 2 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® SA8650P	9.766 ms	0 - 254 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® SA8255P	9.766 ms	0 - 254 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	7.572 ms	0 - 3 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	8.189 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-6690	73.254 ms	1 - 425 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-7790	14.207 ms	0 - 387 MB	NPU
VIT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	3.925 ms	0 - 178 MB	NPU
VIT	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	14.207 ms	0 - 387 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	4.001 ms	0 - 177 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	5.83 ms	0 - 171 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® X2 Elite	4.957 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® X Elite	11.222 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	7.008 ms	0 - 289 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	13.377 ms	0 - 300 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS6490	56.903 ms	0 - 2 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-8275	9.916 ms	0 - 3 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.482 ms	0 - 4 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® QCS8450	13.377 ms	0 - 300 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-9075	9.542 ms	0 - 2 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-X7181	11.222 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-6690	200.814 ms	0 - 519 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-7790	18.779 ms	0 - 302 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-8750	5.83 ms	0 - 171 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® SA7255P	30.811 ms	0 - 171 MB	NPU
VIT	QNN_DLC	w8a8	Qualcomm® SA8295P	16.736 ms	0 - 179 MB	NPU
VIT	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	18.779 ms	0 - 302 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	4.368 ms	0 - 296 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	6.202 ms	0 - 287 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Snapdragon® X2 Elite	15.261 ms	0 - 0 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Snapdragon® X Elite	11.383 ms	1 - 1 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Snapdragon® 8 Gen 3 Mobile	7.345 ms	0 - 405 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ IQ-8275	10.293 ms	0 - 3 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.73 ms	0 - 200 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® SA8650P	11.521 ms	0 - 349 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® SA8255P	11.521 ms	0 - 349 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ IQ-9075	10.031 ms	2 - 4 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ IQ-X7181	11.383 ms	1 - 1 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ Q-6690	227.257 ms	0 - 528 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ Q-7790	20.616 ms	0 - 314 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® Dragonwing™ Q-8750	6.202 ms	0 - 287 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Qualcomm® SA7255P	30.873 ms	0 - 286 MB	NPU
VIT	QNN_DLC	w8a8_mixed_int16	Snapdragon® 7 Gen 4 Mobile	20.616 ms	0 - 314 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	3.611 ms	0 - 162 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	4.753 ms	0 - 160 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	6.755 ms	0 - 290 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	14.545 ms	0 - 286 MB	NPU
VIT	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	13.073 ms	0 - 174 MB	NPU
VIT	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	9.246 ms	0 - 2 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8650P	12.461 ms	0 - 158 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8255P	12.461 ms	0 - 158 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8450	14.545 ms	0 - 286 MB	NPU
VIT	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	12.364 ms	0 - 173 MB	NPU
VIT	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	4.753 ms	0 - 160 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8295P	15.491 ms	0 - 150 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	4.943 ms	0 - 292 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	7.357 ms	0 - 284 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	9.011 ms	0 - 417 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	17.73 ms	0 - 424 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS6490	147.943 ms	1 - 101 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-8275	12.763 ms	0 - 88 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	12.653 ms	0 - 2 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8650P	13.123 ms	0 - 311 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8255P	13.123 ms	0 - 311 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8450	17.73 ms	0 - 424 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-9075	12.488 ms	0 - 87 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-6690	198.344 ms	1 - 269 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-7790	30.129 ms	2 - 173 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-8750	7.357 ms	0 - 284 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA7255P	35.967 ms	0 - 286 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8295P	20.424 ms	0 - 296 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	30.129 ms	2 - 173 MB	NPU

License

The license for the original implementation of VIT 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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/VIT

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Paper • 2010.11929 • Published Oct 22, 2020 • 24