MiniLM-v2: Optimized for Qualcomm Devices

All-MiniLM-L6-v2 maps sentences to a 384-dimensional dense vector space. Trained on 1B+ sentence pairs, it excels at semantic search, clustering, and sentence similarity tasks while being small enough to run on mobile devices.

This is based on the implementation of MiniLM-v2 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
ONNX	w8a8	Universal	QAIRT 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
QNN_DLC	w8a8	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download
TFLITE	w8a8	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit MiniLM-v2 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 MiniLM-v2 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

• Model checkpoint: sentence-transformers/all-MiniLM-L6-v2
• Input resolution: 128 tokens
• Number of parameters: 22.7M
• Model size (float): 86.7 MB
• Embedding dimension: 384

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
MiniLM-v2	ONNX	float	Snapdragon® X2 Elite	0.858 ms	1 - 1 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® X Elite	1.748 ms	45 - 45 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.23 ms	0 - 103 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.536 ms	0 - 108 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.706 ms	0 - 50 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® QCS8450	2.536 ms	0 - 108 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	2.068 ms	0 - 3 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	1.748 ms	45 - 45 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® Dragonwing™ Q-8750	0.862 ms	0 - 61 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Elite Mobile	0.862 ms	0 - 61 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.705 ms	0 - 63 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® X2 Elite	0.672 ms	1 - 1 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® X Elite	1.775 ms	23 - 23 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	1.203 ms	0 - 88 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	1.973 ms	0 - 90 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® Dragonwing™ QCS6490	5.291 ms	0 - 3 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.701 ms	0 - 26 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS8450	1.973 ms	0 - 90 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-9075	1.8 ms	0 - 3 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-X7181	1.775 ms	23 - 23 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® Dragonwing™ Q-8750	0.843 ms	0 - 59 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Elite Mobile	0.843 ms	0 - 59 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.669 ms	0 - 61 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® X2 Elite	0.655 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® X Elite	1.305 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.8 ms	0 - 101 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	1.89 ms	0 - 100 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	3.57 ms	0 - 58 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.12 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA8775P	1.539 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA8650P	1.539 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA8255P	1.539 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS8450	1.89 ms	0 - 100 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	1.387 ms	2 - 4 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	1.305 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	0.565 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA7255P	3.57 ms	0 - 58 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA8295P	2.278 ms	0 - 56 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Elite Mobile	0.565 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.457 ms	0 - 62 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.81 ms	0 - 0 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® X Elite	1.92 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	1.216 ms	0 - 73 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	1.958 ms	0 - 72 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS6490	4.019 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8275	3.416 ms	0 - 47 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.712 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA8775P	1.951 ms	0 - 52 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA8650P	1.951 ms	0 - 52 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA8255P	1.951 ms	0 - 52 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS8450	1.958 ms	0 - 72 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-9075	2.322 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-X7181	1.92 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-6690	6.097 ms	0 - 50 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-7790	1.699 ms	0 - 47 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-8750	0.835 ms	0 - 53 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA7255P	3.416 ms	0 - 47 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA8295P	2.377 ms	0 - 46 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	0.835 ms	0 - 53 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.684 ms	0 - 54 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	1.699 ms	0 - 47 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.834 ms	0 - 104 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	1.952 ms	0 - 102 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	3.684 ms	0 - 63 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.118 ms	0 - 50 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8775P	1.562 ms	0 - 65 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8650P	1.562 ms	0 - 65 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8255P	1.562 ms	0 - 65 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® QCS8450	1.952 ms	0 - 102 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	1.414 ms	0 - 45 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	0.593 ms	0 - 58 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA7255P	3.684 ms	0 - 63 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8295P	2.353 ms	0 - 56 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Elite Mobile	0.593 ms	0 - 58 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.482 ms	0 - 58 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	1.672 ms	0 - 86 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	2.615 ms	0 - 80 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS6490	12.43 ms	0 - 30 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS8275	4.628 ms	0 - 55 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	2.373 ms	0 - 4 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8775P	2.686 ms	0 - 59 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8650P	2.686 ms	0 - 59 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8255P	2.686 ms	0 - 59 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS8450	2.615 ms	0 - 80 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-9075	2.504 ms	0 - 24 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-6690	12.027 ms	0 - 37 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-7790	5.019 ms	0 - 35 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-8750	1.135 ms	0 - 61 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA7255P	4.628 ms	0 - 55 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8295P	3.216 ms	0 - 48 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	1.135 ms	0 - 61 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.95 ms	0 - 53 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	5.019 ms	0 - 35 MB	NPU

License

• The license for the original implementation of MiniLM-v2 can be found here.

References

• Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
• 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.