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 2.42, ONNX Runtime 1.24.1	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
ONNX	w8a8_mixed_int16	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a16	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.17.0	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® 8 Elite Gen 5 Mobile	3.667 ms	1 - 355 MB	NPU
VIT	ONNX	float	Snapdragon® X2 Elite	3.876 ms	170 - 170 MB	NPU
VIT	ONNX	float	Snapdragon® X Elite	11.109 ms	170 - 170 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	7.167 ms	0 - 377 MB	NPU
VIT	ONNX	float	Qualcomm® QCS8550 (Proxy)	10.487 ms	0 - 195 MB	NPU
VIT	ONNX	float	Qualcomm® QCS9075	14.321 ms	0 - 4 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	4.987 ms	0 - 344 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.797 ms	0 - 301 MB	NPU
VIT	ONNX	w8a16	Snapdragon® X2 Elite	3.894 ms	86 - 86 MB	NPU
VIT	ONNX	w8a16	Snapdragon® X Elite	11.261 ms	86 - 86 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	7.372 ms	0 - 379 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCS6490	1108.405 ms	35 - 60 MB	CPU
VIT	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	10.713 ms	0 - 477 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCS9075	12.785 ms	0 - 3 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCM6690	614.255 ms	73 - 88 MB	CPU
VIT	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	5.306 ms	0 - 295 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	594.848 ms	86 - 107 MB	CPU
VIT	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	4.581 ms	0 - 351 MB	NPU
VIT	ONNX	w8a8	Snapdragon® X2 Elite	5.106 ms	85 - 85 MB	NPU
VIT	ONNX	w8a8	Snapdragon® X Elite	13.631 ms	85 - 85 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	8.8 ms	0 - 466 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCS6490	354.711 ms	21 - 79 MB	CPU
VIT	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	12.932 ms	0 - 100 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCS9075	13.588 ms	0 - 3 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCM6690	135.114 ms	14 - 32 MB	CPU
VIT	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	7.277 ms	0 - 319 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	128.787 ms	24 - 43 MB	CPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Elite Gen 5 Mobile	63.145 ms	0 - 268 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® X2 Elite	53.502 ms	79 - 79 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® X Elite	174.304 ms	79 - 79 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Gen 3 Mobile	81.808 ms	67 - 429 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS6490	728.091 ms	97 - 126 MB	CPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS8550 (Proxy)	101.273 ms	0 - 382 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS9075	130.774 ms	68 - 70 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCM6690	387.043 ms	103 - 124 MB	CPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	88.161 ms	68 - 340 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 7 Gen 4 Mobile	372.061 ms	35 - 55 MB	CPU
VIT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	4.052 ms	1 - 344 MB	NPU
VIT	QNN_DLC	float	Snapdragon® X2 Elite	4.478 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Snapdragon® X Elite	11.859 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	7.716 ms	0 - 369 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	40.633 ms	1 - 339 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	11.125 ms	1 - 407 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8775P	64.028 ms	1 - 339 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS9075	14.893 ms	1 - 3 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	19.096 ms	0 - 350 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA7255P	40.633 ms	1 - 339 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8295P	17.182 ms	1 - 332 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	5.295 ms	0 - 338 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	3.094 ms	0 - 278 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	5.87 ms	0 - 318 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8275 (Proxy)	35.876 ms	0 - 288 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8550 (Proxy)	8.002 ms	0 - 3 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8775P	11.156 ms	0 - 288 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS9075	11.649 ms	0 - 174 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8450 (Proxy)	13.857 ms	0 - 292 MB	NPU
VIT	TFLITE	float	Qualcomm® SA7255P	35.876 ms	0 - 288 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8295P	13.383 ms	0 - 262 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	3.943 ms	0 - 291 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	2.295 ms	0 - 88 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	4.735 ms	0 - 182 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS6490	78.182 ms	1 - 99 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	14.341 ms	0 - 85 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	6.776 ms	0 - 4 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8775P	7.074 ms	0 - 86 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS9075	7.561 ms	0 - 89 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCM6690	101.681 ms	1 - 185 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	8.676 ms	0 - 181 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA7255P	14.341 ms	0 - 85 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8295P	9.695 ms	0 - 89 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	3.374 ms	0 - 84 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	20.157 ms	1 - 74 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 • 15