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 1.24.3	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
ONNX	w8a8_mixed_int16	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.19.1	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	2.698 ms	1 - 347 MB	NPU
VIT	ONNX	float	Snapdragon® X2 Elite	3.028 ms	170 - 170 MB	NPU
VIT	ONNX	float	Snapdragon® X Elite	7.669 ms	170 - 170 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	5.03 ms	0 - 369 MB	NPU
VIT	ONNX	float	Qualcomm® QCS8550 (Proxy)	7.131 ms	0 - 196 MB	NPU
VIT	ONNX	float	Qualcomm® QCS9075	10.169 ms	0 - 4 MB	NPU
VIT	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	3.44 ms	0 - 339 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.116 ms	0 - 276 MB	NPU
VIT	ONNX	w8a16	Snapdragon® X2 Elite	4.424 ms	86 - 86 MB	NPU
VIT	ONNX	w8a16	Snapdragon® X Elite	8.794 ms	86 - 86 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	5.734 ms	0 - 349 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCS6490	1125.779 ms	31 - 54 MB	CPU
VIT	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	8.19 ms	0 - 102 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCS9075	8.771 ms	0 - 3 MB	NPU
VIT	ONNX	w8a16	Qualcomm® QCM6690	604.448 ms	54 - 70 MB	CPU
VIT	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	4.395 ms	0 - 281 MB	NPU
VIT	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	590.731 ms	42 - 59 MB	CPU
VIT	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	4.576 ms	0 - 421 MB	NPU
VIT	ONNX	w8a8	Snapdragon® X2 Elite	5.214 ms	85 - 85 MB	NPU
VIT	ONNX	w8a8	Snapdragon® X Elite	13.707 ms	85 - 85 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	8.762 ms	0 - 530 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCS6490	308.621 ms	19 - 72 MB	CPU
VIT	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	12.993 ms	0 - 101 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCS9075	13.737 ms	0 - 3 MB	NPU
VIT	ONNX	w8a8	Qualcomm® QCM6690	131.058 ms	22 - 41 MB	CPU
VIT	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	6.966 ms	0 - 413 MB	NPU
VIT	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	125.982 ms	11 - 31 MB	CPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® X2 Elite	45.84 ms	75 - 75 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® X Elite	160.974 ms	73 - 73 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Gen 3 Mobile	74.098 ms	54 - 418 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS6490	729.381 ms	77 - 111 MB	CPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS8550 (Proxy)	93.613 ms	51 - 57 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCS9075	115.883 ms	55 - 57 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Qualcomm® QCM6690	363.109 ms	41 - 61 MB	CPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	63.713 ms	51 - 335 MB	NPU
VIT	ONNX	w8a8_mixed_int16	Snapdragon® 7 Gen 4 Mobile	352.78 ms	81 - 104 MB	CPU
VIT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.993 ms	0 - 318 MB	NPU
VIT	QNN_DLC	float	Snapdragon® X2 Elite	3.572 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Snapdragon® X Elite	8.504 ms	1 - 1 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	5.519 ms	0 - 357 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	35.226 ms	1 - 331 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	7.912 ms	1 - 2 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8775P	10.654 ms	1 - 332 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS9075	11.097 ms	1 - 3 MB	NPU
VIT	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	12.772 ms	0 - 337 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA7255P	35.226 ms	1 - 331 MB	NPU
VIT	QNN_DLC	float	Qualcomm® SA8295P	13.047 ms	1 - 310 MB	NPU
VIT	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	3.759 ms	0 - 333 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.955 ms	0 - 275 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	5.276 ms	0 - 355 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8275 (Proxy)	34.552 ms	0 - 276 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8550 (Proxy)	7.191 ms	0 - 3 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8775P	10.187 ms	0 - 277 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS9075	10.641 ms	0 - 174 MB	NPU
VIT	TFLITE	float	Qualcomm® QCS8450 (Proxy)	12.92 ms	0 - 339 MB	NPU
VIT	TFLITE	float	Qualcomm® SA7255P	34.552 ms	0 - 276 MB	NPU
VIT	TFLITE	float	Qualcomm® SA8295P	13.247 ms	0 - 265 MB	NPU
VIT	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	3.701 ms	0 - 285 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	4.525 ms	0 - 381 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	8.662 ms	0 - 493 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS6490	166.626 ms	1 - 101 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	35.043 ms	0 - 386 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	12.909 ms	0 - 2 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8775P	12.203 ms	0 - 388 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS9075	13.759 ms	0 - 88 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCM6690	199.256 ms	2 - 293 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	20.024 ms	0 - 456 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA7255P	35.043 ms	0 - 386 MB	NPU
VIT	TFLITE	w8a8	Qualcomm® SA8295P	19.189 ms	0 - 344 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	6.932 ms	0 - 381 MB	NPU
VIT	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	29.256 ms	2 - 222 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