ConvNext-Tiny: Optimized for Qualcomm Devices

ConvNextTiny 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 ConvNext-Tiny 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 2.37, ONNX Runtime 1.23.0	Download
ONNX	w8a16	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit ConvNext-Tiny 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 ConvNext-Tiny 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: 28.6M
• Model size (float): 109 MB
• Model size (w8a16): 28.9 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ConvNext-Tiny	ONNX	float	Snapdragon® X Elite	2.871 ms	57 - 57 MB	NPU
ConvNext-Tiny	ONNX	float	Snapdragon® 8 Gen 3 Mobile	2.254 ms	0 - 227 MB	NPU
ConvNext-Tiny	ONNX	float	Qualcomm® QCS8550 (Proxy)	3.041 ms	0 - 66 MB	NPU
ConvNext-Tiny	ONNX	float	Qualcomm® QCS9075	4.152 ms	1 - 4 MB	NPU
ConvNext-Tiny	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.798 ms	0 - 181 MB	NPU
ConvNext-Tiny	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.465 ms	1 - 182 MB	NPU
ConvNext-Tiny	ONNX	w8a16	Snapdragon® X Elite	66.853 ms	62 - 62 MB	NPU
ConvNext-Tiny	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	88.472 ms	45 - 194 MB	NPU
ConvNext-Tiny	ONNX	w8a16	Qualcomm® QCS6490	397.835 ms	50 - 69 MB	CPU
ConvNext-Tiny	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	99.954 ms	42 - 45 MB	NPU
ConvNext-Tiny	ONNX	w8a16	Qualcomm® QCS9075	105.638 ms	48 - 50 MB	NPU
ConvNext-Tiny	ONNX	w8a16	Qualcomm® QCM6690	244.731 ms	59 - 72 MB	CPU
ConvNext-Tiny	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	76.665 ms	45 - 162 MB	NPU
ConvNext-Tiny	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	231.806 ms	57 - 71 MB	CPU
ConvNext-Tiny	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	70.719 ms	42 - 160 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Snapdragon® X Elite	3.932 ms	1 - 1 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	2.665 ms	1 - 170 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	15.324 ms	1 - 124 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	3.68 ms	1 - 2 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® SA8775P	5.008 ms	1 - 126 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® QCS9075	4.856 ms	1 - 3 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	9.628 ms	0 - 168 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® SA7255P	15.324 ms	1 - 124 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Qualcomm® SA8295P	8.903 ms	1 - 125 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	2.028 ms	0 - 126 MB	NPU
ConvNext-Tiny	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.614 ms	1 - 127 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Snapdragon® X Elite	3.411 ms	0 - 0 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	2.17 ms	0 - 121 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® QCS6490	9.087 ms	0 - 2 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	6.814 ms	0 - 96 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	3.135 ms	0 - 2 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® SA8775P	3.46 ms	0 - 97 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® QCS9075	3.331 ms	0 - 2 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® QCM6690	23.577 ms	0 - 250 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	4.17 ms	0 - 124 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® SA7255P	6.814 ms	0 - 96 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Qualcomm® SA8295P	4.656 ms	0 - 97 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	1.591 ms	0 - 101 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	3.428 ms	0 - 108 MB	NPU
ConvNext-Tiny	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	1.286 ms	0 - 100 MB	NPU
ConvNext-Tiny	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	2.204 ms	0 - 169 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® QCS8275 (Proxy)	14.244 ms	0 - 122 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® QCS8550 (Proxy)	2.913 ms	0 - 2 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® SA8775P	4.271 ms	0 - 123 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® QCS9075	4.082 ms	0 - 59 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® QCS8450 (Proxy)	8.877 ms	0 - 161 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® SA7255P	14.244 ms	0 - 122 MB	NPU
ConvNext-Tiny	TFLITE	float	Qualcomm® SA8295P	7.835 ms	0 - 119 MB	NPU
ConvNext-Tiny	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.643 ms	0 - 127 MB	NPU
ConvNext-Tiny	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.333 ms	0 - 121 MB	NPU

License

• The license for the original implementation of ConvNext-Tiny can be found here.

References

• A ConvNet for the 2020s
• 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.