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 1.25.0	Download
ONNX	w8a8	Universal	QAIRT 2.45, ONNX Runtime 1.25.0	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	0.774 ms	212 - 212 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® X Elite	1.769 ms	149 - 149 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.231 ms	0 - 93 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.535 ms	0 - 98 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.772 ms	0 - 3 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® QCS8450	2.535 ms	0 - 98 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Elite Mobile	0.866 ms	0 - 62 MB	NPU
MiniLM-v2	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.698 ms	0 - 58 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® QCS9075	2.964 ms	0 - 45 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® QCS8750	0.866 ms	0 - 62 MB	NPU
MiniLM-v2	ONNX	float	Qualcomm® QCS7181	1.769 ms	149 - 149 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® X2 Elite	0.669 ms	213 - 213 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® X Elite	1.788 ms	149 - 149 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	1.194 ms	0 - 78 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	1.98 ms	0 - 78 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS6490	4.839 ms	0 - 45 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	1.716 ms	0 - 27 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS8450	1.98 ms	0 - 78 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.664 ms	0 - 56 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	1.943 ms	0 - 51 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCM6690	6.704 ms	0 - 57 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS9075	1.871 ms	0 - 45 MB	NPU
MiniLM-v2	ONNX	w8a8	Snapdragon® 8 Elite Mobile	0.84 ms	0 - 56 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS7790	1.943 ms	0 - 51 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS8750	0.84 ms	0 - 56 MB	NPU
MiniLM-v2	ONNX	w8a8	Qualcomm® QCS7181	1.788 ms	149 - 149 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® X2 Elite	0.682 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® X Elite	1.302 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.795 ms	0 - 101 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	1.886 ms	0 - 100 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS8275	3.561 ms	0 - 58 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.125 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS8450	1.886 ms	0 - 100 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Elite Mobile	0.567 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA8295P	2.301 ms	0 - 56 MB	NPU
MiniLM-v2	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.461 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® SA7255P	3.561 ms	0 - 58 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS9075	1.451 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS8750	0.567 ms	0 - 60 MB	NPU
MiniLM-v2	QNN_DLC	float	Qualcomm® QCS7181	1.302 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.819 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® X Elite	1.912 ms	1 - 1 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	1.206 ms	0 - 69 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	1.952 ms	0 - 72 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS6490	3.699 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS8275	3.456 ms	0 - 48 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	1.715 ms	0 - 51 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS8450	1.952 ms	0 - 72 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.681 ms	0 - 54 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	1.694 ms	0 - 47 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCM6690	6.088 ms	2 - 52 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS9075	1.862 ms	0 - 2 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA7255P	3.456 ms	0 - 48 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	0.839 ms	0 - 52 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® SA8295P	2.372 ms	0 - 46 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS7790	1.694 ms	0 - 47 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS8750	0.839 ms	0 - 52 MB	NPU
MiniLM-v2	QNN_DLC	w8a8	Qualcomm® QCS7181	1.912 ms	1 - 1 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.832 ms	0 - 102 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	1.932 ms	0 - 102 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® QCS8275	3.674 ms	0 - 63 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.127 ms	0 - 2 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8775P	12.459 ms	0 - 28 MB	GPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8650P	12.459 ms	0 - 28 MB	GPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8255P	12.459 ms	0 - 28 MB	GPU
MiniLM-v2	TFLITE	float	Qualcomm® QCS8450	1.932 ms	0 - 102 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Elite Mobile	0.593 ms	0 - 65 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA8295P	2.366 ms	0 - 56 MB	NPU
MiniLM-v2	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.482 ms	0 - 58 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® SA7255P	3.674 ms	0 - 63 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® QCS9075	1.501 ms	0 - 45 MB	NPU
MiniLM-v2	TFLITE	float	Qualcomm® QCS8750	0.593 ms	0 - 65 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	1.672 ms	0 - 83 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	2.619 ms	0 - 80 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS6490	13.129 ms	0 - 31 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS8275	4.618 ms	0 - 55 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	2.366 ms	0 - 3 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8775P	12.541 ms	0 - 30 MB	GPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8650P	12.541 ms	0 - 30 MB	GPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8255P	12.541 ms	0 - 30 MB	GPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS8450	2.619 ms	0 - 80 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.947 ms	0 - 54 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	5.123 ms	0 - 35 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCM6690	12.046 ms	0 - 35 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS9075	2.498 ms	0 - 24 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA7255P	4.618 ms	0 - 55 MB	NPU
MiniLM-v2	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	1.134 ms	0 - 62 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® SA8295P	3.202 ms	0 - 48 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS7790	5.123 ms	0 - 35 MB	NPU
MiniLM-v2	TFLITE	w8a8	Qualcomm® QCS8750	1.134 ms	0 - 62 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.