Albert-Base-V2-Hf: Optimized for Qualcomm Devices

ALBERT 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 Albert-Base-V2-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	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 Albert-Base-V2-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 Albert-Base-V2-Hf on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

• Model checkpoint: albert/albert-base-v2
• Input resolution: 1x384
• Number of parameters: 11.8M
• Model size (float): 43.9 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Albert-Base-V2-Hf	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	12.516 ms	0 - 411 MB	NPU
Albert-Base-V2-Hf	ONNX	float	Snapdragon® X2 Elite	14.864 ms	32 - 32 MB	NPU
Albert-Base-V2-Hf	ONNX	float	Snapdragon® X Elite	27.082 ms	32 - 32 MB	NPU
Albert-Base-V2-Hf	ONNX	float	Snapdragon® 8 Gen 3 Mobile	20.554 ms	0 - 455 MB	NPU
Albert-Base-V2-Hf	ONNX	float	Qualcomm® QCS8550 (Proxy)	27.785 ms	0 - 43 MB	NPU
Albert-Base-V2-Hf	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	15.655 ms	0 - 377 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	8.524 ms	0 - 291 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Snapdragon® X2 Elite	8.633 ms	22 - 22 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Snapdragon® X Elite	19.918 ms	22 - 22 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	14.172 ms	0 - 375 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Qualcomm® QCS6490	2214.118 ms	94 - 121 MB	CPU
Albert-Base-V2-Hf	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	19.771 ms	0 - 31 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Qualcomm® QCS9075	21.234 ms	0 - 3 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Qualcomm® QCM6690	1152.49 ms	108 - 123 MB	CPU
Albert-Base-V2-Hf	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	10.693 ms	0 - 302 MB	NPU
Albert-Base-V2-Hf	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	1110.622 ms	82 - 94 MB	CPU
Albert-Base-V2-Hf	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	9.731 ms	0 - 396 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Snapdragon® X2 Elite	10.552 ms	1 - 1 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Snapdragon® X Elite	22.548 ms	0 - 0 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	17.206 ms	0 - 372 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	75.133 ms	0 - 316 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	22.835 ms	0 - 2 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® SA8775P	27.641 ms	0 - 315 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® QCS9075	26.26 ms	0 - 2 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	51.357 ms	0 - 420 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® SA7255P	75.133 ms	0 - 316 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Qualcomm® SA8295P	34.424 ms	0 - 384 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	11.82 ms	0 - 386 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	5.486 ms	0 - 257 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Snapdragon® X2 Elite	6.204 ms	0 - 0 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Snapdragon® X Elite	13.74 ms	0 - 0 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	9.575 ms	0 - 289 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	29.762 ms	0 - 249 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	13.293 ms	0 - 2 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Qualcomm® SA8775P	60.357 ms	0 - 249 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Qualcomm® QCS9075	15.59 ms	0 - 2 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Qualcomm® SA7255P	29.762 ms	0 - 249 MB	NPU
Albert-Base-V2-Hf	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	7.77 ms	0 - 271 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	10.077 ms	0 - 402 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	17.501 ms	0 - 389 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® QCS8275 (Proxy)	74.887 ms	0 - 330 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® QCS8550 (Proxy)	22.888 ms	0 - 3 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® SA8775P	27.347 ms	0 - 398 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® QCS9075	26.947 ms	0 - 33 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® QCS8450 (Proxy)	35.921 ms	0 - 425 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® SA7255P	74.887 ms	0 - 330 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Qualcomm® SA8295P	34.735 ms	0 - 383 MB	NPU
Albert-Base-V2-Hf	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	12.109 ms	0 - 383 MB	NPU

License

• The license for the original implementation of Albert-Base-V2-Hf can be found here.

References

• ALBERT: A Lite BERT for Self-supervised Learning of Language Representations
• 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.