Bert-Base-Uncased-Hf: Optimized for Qualcomm Devices

Bert is a lightweight BERT model designed for efficient self-supervised learning of language representations. It can be used for masked language modeling and as a backbone for various NLP tasks.

This is based on the implementation of Bert-Base-Uncased-Hf 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	w8a16	Universal	QAIRT 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
QNN_DLC	w8a16	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit Bert-Base-Uncased-Hf 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 Bert-Base-Uncased-Hf on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

Input resolution: 1x384
Model checkpoint: google-bert/bert-base-uncased
Model size (float): 418 MB
Number of parameters: 110M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® X2 Elite	5.463 ms	1 - 1 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® X Elite	15.642 ms	154 - 154 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	11.464 ms	0 - 525 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® 8 Gen 1 Mobile	24.541 ms	0 - 527 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ QCS6490	118.221 ms	60 - 63 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-8275	14.929 ms	0 - 4 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	15.66 ms	0 - 163 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® QCS8450	24.541 ms	0 - 527 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	15.499 ms	0 - 3 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	15.642 ms	154 - 154 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ Q-6690	170.984 ms	58 - 595 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	35.154 ms	59 - 538 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	8.845 ms	0 - 460 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® 8 Elite Mobile	8.845 ms	0 - 460 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	4.906 ms	0 - 466 MB	NPU
Bert-Base-Uncased-Hf	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	35.154 ms	59 - 538 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Snapdragon® X2 Elite	21.637 ms	1 - 1 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Snapdragon® X Elite	18.778 ms	0 - 0 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	12.42 ms	0 - 540 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	36.531 ms	0 - 575 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	24.617 ms	0 - 2 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	17.754 ms	0 - 2 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® SA8775P	24.295 ms	0 - 358 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® SA8650P	24.295 ms	0 - 358 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® SA8255P	24.295 ms	0 - 358 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® QCS8450	36.531 ms	0 - 575 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	24.159 ms	0 - 2 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	18.778 ms	0 - 0 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	9.051 ms	0 - 395 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® SA7255P	76.844 ms	0 - 358 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Qualcomm® SA8295P	30.133 ms	0 - 397 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Snapdragon® 8 Elite Mobile	9.051 ms	0 - 395 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	6.61 ms	0 - 346 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Snapdragon® X2 Elite	5.426 ms	1 - 1 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Snapdragon® X Elite	12.107 ms	0 - 0 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	7.849 ms	0 - 519 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-8275	10.561 ms	0 - 2 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.424 ms	0 - 2 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® SA8775P	11.157 ms	0 - 446 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® SA8650P	11.157 ms	0 - 446 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® SA8255P	11.157 ms	0 - 446 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	11.044 ms	2 - 4 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	12.107 ms	0 - 0 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	6.429 ms	0 - 458 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Qualcomm® SA7255P	27.074 ms	0 - 444 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Snapdragon® 8 Elite Mobile	6.429 ms	0 - 458 MB	NPU
Bert-Base-Uncased-Hf	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	4.359 ms	0 - 463 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	12.499 ms	0 - 555 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	36.33 ms	0 - 584 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	24.758 ms	0 - 260 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	17.143 ms	0 - 2 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® SA8775P	24.491 ms	0 - 373 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® SA8650P	24.491 ms	0 - 373 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® SA8255P	24.491 ms	0 - 373 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® QCS8450	36.33 ms	0 - 584 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	24.672 ms	0 - 259 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	9.202 ms	0 - 404 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® SA7255P	77.276 ms	0 - 373 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Qualcomm® SA8295P	30.149 ms	0 - 400 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Snapdragon® 8 Elite Mobile	9.202 ms	0 - 404 MB	NPU
Bert-Base-Uncased-Hf	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	6.864 ms	0 - 387 MB	NPU

License

The license for the original implementation of Bert-Base-Uncased-Hf can be found here.

References

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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/Bert-Base-Uncased-Hf

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Paper • 1810.04805 • Published Oct 11, 2018 • 33