Beit: Optimized for Qualcomm Devices

Beit 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 Beit 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
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a16	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 Beit 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 Beit 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: 92.0M
• Model size (float): 351 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Beit	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	6.237 ms	1 - 486 MB	NPU
Beit	ONNX	float	Snapdragon® X2 Elite	6.019 ms	185 - 185 MB	NPU
Beit	ONNX	float	Snapdragon® X Elite	13.686 ms	185 - 185 MB	NPU
Beit	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.267 ms	0 - 527 MB	NPU
Beit	ONNX	float	Qualcomm® QCS8550 (Proxy)	12.928 ms	0 - 195 MB	NPU
Beit	ONNX	float	Qualcomm® QCS9075	17.598 ms	0 - 4 MB	NPU
Beit	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	6.65 ms	1 - 493 MB	NPU
Beit	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	4.182 ms	0 - 407 MB	NPU
Beit	ONNX	w8a16	Snapdragon® X2 Elite	4.371 ms	96 - 96 MB	NPU
Beit	ONNX	w8a16	Snapdragon® X Elite	12.511 ms	96 - 96 MB	NPU
Beit	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	7.927 ms	0 - 492 MB	NPU
Beit	ONNX	w8a16	Qualcomm® QCS6490	1060.589 ms	53 - 70 MB	CPU
Beit	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	11.769 ms	0 - 6 MB	NPU
Beit	ONNX	w8a16	Qualcomm® QCS9075	14.665 ms	0 - 3 MB	NPU
Beit	ONNX	w8a16	Qualcomm® QCM6690	599.901 ms	112 - 128 MB	CPU
Beit	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	6.102 ms	0 - 408 MB	NPU
Beit	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	582.769 ms	73 - 87 MB	CPU
Beit	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	6.571 ms	1 - 481 MB	NPU
Beit	QNN_DLC	float	Snapdragon® X2 Elite	6.925 ms	1 - 1 MB	NPU
Beit	QNN_DLC	float	Snapdragon® X Elite	13.429 ms	1 - 1 MB	NPU
Beit	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.695 ms	0 - 535 MB	NPU
Beit	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	44.932 ms	1 - 485 MB	NPU
Beit	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	12.588 ms	1 - 3 MB	NPU
Beit	QNN_DLC	float	Qualcomm® SA8775P	16.39 ms	1 - 485 MB	NPU
Beit	QNN_DLC	float	Qualcomm® QCS9075	16.723 ms	1 - 3 MB	NPU
Beit	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	22.928 ms	1 - 509 MB	NPU
Beit	QNN_DLC	float	Qualcomm® SA7255P	44.932 ms	1 - 485 MB	NPU
Beit	QNN_DLC	float	Qualcomm® SA8295P	19.07 ms	1 - 468 MB	NPU
Beit	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	6.965 ms	1 - 480 MB	NPU
Beit	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	3.971 ms	0 - 297 MB	NPU
Beit	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	6.668 ms	0 - 343 MB	NPU
Beit	TFLITE	float	Qualcomm® QCS8275 (Proxy)	38.591 ms	0 - 297 MB	NPU
Beit	TFLITE	float	Qualcomm® QCS8550 (Proxy)	9.334 ms	0 - 3 MB	NPU
Beit	TFLITE	float	Qualcomm® SA8775P	55.63 ms	0 - 305 MB	NPU
Beit	TFLITE	float	Qualcomm® QCS9075	13.213 ms	0 - 187 MB	NPU
Beit	TFLITE	float	Qualcomm® QCS8450 (Proxy)	19.184 ms	0 - 430 MB	NPU
Beit	TFLITE	float	Qualcomm® SA7255P	38.591 ms	0 - 297 MB	NPU
Beit	TFLITE	float	Qualcomm® SA8295P	16.061 ms	0 - 405 MB	NPU
Beit	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	4.721 ms	0 - 297 MB	NPU

License

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

References

• BEIT: BERT Pre-Training of Image Transformers
• 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.