EfficientViT-b2-cls: Optimized for Qualcomm Devices

EfficientViT 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 EfficientViT-b2-cls 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	w8a16_mixed_fp16	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit EfficientViT-b2-cls 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 EfficientViT-b2-cls 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: 24.3M
• Model size (float): 92.9 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
EfficientViT-b2-cls	ONNX	float	Snapdragon® X Elite	5.903 ms	49 - 49 MB	NPU
EfficientViT-b2-cls	ONNX	float	Snapdragon® 8 Gen 3 Mobile	3.625 ms	0 - 181 MB	NPU
EfficientViT-b2-cls	ONNX	float	Qualcomm® QCS8550 (Proxy)	5.163 ms	0 - 58 MB	NPU
EfficientViT-b2-cls	ONNX	float	Qualcomm® QCS9075	5.828 ms	1 - 4 MB	NPU
EfficientViT-b2-cls	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	2.693 ms	0 - 89 MB	NPU
EfficientViT-b2-cls	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2.272 ms	0 - 115 MB	NPU
EfficientViT-b2-cls	ONNX	float	Snapdragon® X2 Elite	2.54 ms	49 - 49 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Snapdragon® X Elite	5.981 ms	1 - 1 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	3.777 ms	0 - 164 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	13.008 ms	1 - 90 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	5.352 ms	1 - 215 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Qualcomm® QCS9075	6.201 ms	3 - 5 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	7.193 ms	0 - 164 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	2.79 ms	0 - 91 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.334 ms	1 - 95 MB	NPU
EfficientViT-b2-cls	QNN_DLC	float	Snapdragon® X2 Elite	2.961 ms	1 - 1 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	3.794 ms	0 - 217 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Qualcomm® QCS8275 (Proxy)	13.061 ms	0 - 148 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Qualcomm® QCS8550 (Proxy)	5.352 ms	0 - 3 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Qualcomm® QCS9075	6.232 ms	0 - 52 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Qualcomm® QCS8450 (Proxy)	7.167 ms	0 - 223 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	2.781 ms	0 - 153 MB	NPU
EfficientViT-b2-cls	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.337 ms	0 - 155 MB	NPU

License

• The license for the original implementation of EfficientViT-b2-cls can be found here.

References

• EfficientViT: Multi-Scale Linear Attention for High-Resolution Dense Prediction
• 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.