RepViT: Optimized for Qualcomm Devices

RepViT is a lightweight pure CNN model for mobile devices, incorporating efficient architectural designs from Vision Transformers into CNNs. It achieves over 80% top-1 accuracy on ImageNet with 1ms latency on an iPhone 12.

This is based on the implementation of RepViT 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
QNN_DLC	float	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit RepViT 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 RepViT 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): 91 MB
• Number of parameters: 22.9M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
RepViT	ONNX	float	Snapdragon® X2 Elite	1.703 ms	2 - 2 MB	NPU
RepViT	ONNX	float	Snapdragon® X Elite	3.352 ms	47 - 47 MB	NPU
RepViT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	2.378 ms	0 - 196 MB	NPU
RepViT	ONNX	float	Snapdragon® 8 Gen 1 Mobile	11.525 ms	1 - 193 MB	NPU
RepViT	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	4.617 ms	0 - 4 MB	NPU
RepViT	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	3.182 ms	0 - 232 MB	NPU
RepViT	ONNX	float	Qualcomm® QCS8450	11.525 ms	1 - 193 MB	NPU
RepViT	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	4.872 ms	0 - 4 MB	NPU
RepViT	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	3.352 ms	47 - 47 MB	NPU
RepViT	ONNX	float	Qualcomm® Dragonwing™ Q-8750	1.814 ms	0 - 114 MB	NPU
RepViT	ONNX	float	Snapdragon® 8 Elite Mobile	1.814 ms	0 - 114 MB	NPU
RepViT	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.453 ms	0 - 114 MB	NPU
RepViT	QNN_DLC	float	Snapdragon® X2 Elite	2.103 ms	1 - 1 MB	NPU
RepViT	QNN_DLC	float	Snapdragon® X Elite	3.861 ms	1 - 1 MB	NPU
RepViT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	2.545 ms	0 - 177 MB	NPU
RepViT	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	12.232 ms	1 - 177 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	4.764 ms	1 - 4 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	14.457 ms	1 - 101 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	3.467 ms	1 - 19 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® SA8775P	4.83 ms	1 - 103 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® SA8650P	4.83 ms	1 - 103 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® SA8255P	4.83 ms	1 - 103 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® QCS8450	12.232 ms	1 - 177 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	4.578 ms	3 - 5 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	3.861 ms	1 - 1 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	1.909 ms	0 - 104 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® SA7255P	14.457 ms	1 - 101 MB	NPU
RepViT	QNN_DLC	float	Qualcomm® SA8295P	9.678 ms	1 - 106 MB	NPU
RepViT	QNN_DLC	float	Snapdragon® 8 Elite Mobile	1.909 ms	0 - 104 MB	NPU
RepViT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.508 ms	1 - 104 MB	NPU
RepViT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	2.546 ms	0 - 234 MB	NPU
RepViT	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	12.232 ms	0 - 228 MB	NPU
RepViT	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	4.776 ms	0 - 51 MB	NPU
RepViT	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	14.489 ms	0 - 162 MB	NPU
RepViT	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	3.451 ms	0 - 2 MB	NPU
RepViT	TFLITE	float	Qualcomm® SA8775P	4.868 ms	0 - 164 MB	NPU
RepViT	TFLITE	float	Qualcomm® SA8650P	4.868 ms	0 - 164 MB	NPU
RepViT	TFLITE	float	Qualcomm® SA8255P	4.868 ms	0 - 164 MB	NPU
RepViT	TFLITE	float	Qualcomm® QCS8450	12.232 ms	0 - 228 MB	NPU
RepViT	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	4.588 ms	0 - 50 MB	NPU
RepViT	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	1.89 ms	0 - 164 MB	NPU
RepViT	TFLITE	float	Qualcomm® SA7255P	14.489 ms	0 - 162 MB	NPU
RepViT	TFLITE	float	Qualcomm® SA8295P	9.653 ms	0 - 161 MB	NPU
RepViT	TFLITE	float	Snapdragon® 8 Elite Mobile	1.89 ms	0 - 164 MB	NPU
RepViT	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.508 ms	0 - 167 MB	NPU

License

• The license for the original implementation of RepViT can be found here.

References

• RepViT: Revisiting Mobile CNN From ViT Perspective
• 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.