EfficientViT-l2-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-l2-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 2.45, 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 EfficientViT-l2-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-l2-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: 63.7M
• Model size (float): 243 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
EfficientViT-l2-cls	ONNX	float	Snapdragon® X2 Elite	7.011 ms	2 - 2 MB	NPU
EfficientViT-l2-cls	ONNX	float	Snapdragon® X Elite	14.426 ms	131 - 131 MB	NPU
EfficientViT-l2-cls	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.933 ms	0 - 478 MB	NPU
EfficientViT-l2-cls	ONNX	float	Snapdragon® 8 Gen 1 Mobile	39.976 ms	1 - 356 MB	NPU
EfficientViT-l2-cls	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	14.019 ms	0 - 162 MB	NPU
EfficientViT-l2-cls	ONNX	float	Qualcomm® QCS8450	39.976 ms	1 - 356 MB	NPU
EfficientViT-l2-cls	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	20.731 ms	1 - 6 MB	NPU
EfficientViT-l2-cls	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	14.426 ms	131 - 131 MB	NPU
EfficientViT-l2-cls	ONNX	float	Qualcomm® Dragonwing™ Q-8750	7.711 ms	2 - 258 MB	NPU
EfficientViT-l2-cls	ONNX	float	Snapdragon® 8 Elite Mobile	7.711 ms	2 - 258 MB	NPU
EfficientViT-l2-cls	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.978 ms	2 - 265 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Snapdragon® X2 Elite	8.668 ms	2 - 2 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Snapdragon® X Elite	18.251 ms	2 - 2 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	12.182 ms	1 - 521 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	39.701 ms	0 - 355 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	59.595 ms	2 - 263 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	17.179 ms	2 - 4 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® SA8775P	21.71 ms	2 - 263 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® SA8650P	21.71 ms	2 - 263 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® SA8255P	21.71 ms	2 - 263 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® QCS8450	39.701 ms	0 - 355 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	24.321 ms	4 - 7 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	18.251 ms	2 - 2 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	8.892 ms	2 - 257 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® SA7255P	59.595 ms	2 - 263 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Qualcomm® SA8295P	34.748 ms	2 - 251 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Snapdragon® 8 Elite Mobile	8.892 ms	2 - 257 MB	NPU
EfficientViT-l2-cls	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	6.96 ms	2 - 267 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	11.742 ms	0 - 609 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	41.61 ms	0 - 448 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	58.377 ms	0 - 468 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	16.287 ms	0 - 3 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® SA8775P	20.811 ms	0 - 470 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® SA8650P	20.811 ms	0 - 470 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® SA8255P	20.811 ms	0 - 470 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® QCS8450	41.61 ms	0 - 448 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	21.067 ms	0 - 135 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	8.665 ms	0 - 334 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® SA7255P	58.377 ms	0 - 468 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Qualcomm® SA8295P	34.286 ms	0 - 321 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Snapdragon® 8 Elite Mobile	8.665 ms	0 - 334 MB	NPU
EfficientViT-l2-cls	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	6.82 ms	0 - 352 MB	NPU

License

• The license for the original implementation of EfficientViT-l2-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.