README.md

---
library_name: pytorch
license: other
tags:
- backbone
- android
pipeline_tag: image-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/beit/web-assets/model_demo.png)

# Beit: Optimized for Qualcomm Devices

Beit 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 Beit found [here](https://github.com/microsoft/unilm/tree/master/beit).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/beit) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) 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](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/beit/releases/v0.59.0/beit-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/beit/releases/v0.59.0/beit-onnx-w8a16.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/beit/releases/v0.59.0/beit-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/beit/releases/v0.59.0/beit-qnn_dlc-w8a16.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/beit/releases/v0.59.0/beit-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[Beit on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/beit)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/beit) 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 [Beit on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/beit) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 92.0M
- Model size (float): 351 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Beit | ONNX | float | Snapdragon® X2 Elite | 7.436 ms | 2 - 2 MB | NPU
| Beit | ONNX | float | Snapdragon® X Elite | 15.305 ms | 184 - 184 MB | NPU
| Beit | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 10.496 ms | 0 - 449 MB | NPU
| Beit | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 17.894 ms | 1 - 422 MB | NPU
| Beit | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 14.88 ms | 0 - 194 MB | NPU
| Beit | ONNX | float | Qualcomm® QCS8450 | 17.894 ms | 1 - 422 MB | NPU
| Beit | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 17.267 ms | 0 - 4 MB | NPU
| Beit | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 15.305 ms | 184 - 184 MB | NPU
| Beit | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 8.532 ms | 1 - 305 MB | NPU
| Beit | ONNX | float | Snapdragon® 8 Elite Mobile | 8.532 ms | 1 - 305 MB | NPU
| Beit | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.161 ms | 0 - 305 MB | NPU
| Beit | ONNX | w8a16 | Snapdragon® X2 Elite | 2.67 ms | 1 - 1 MB | NPU
| Beit | ONNX | w8a16 | Snapdragon® X Elite | 6.899 ms | 96 - 96 MB | NPU
| Beit | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 4.572 ms | 0 - 422 MB | NPU
| Beit | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 11.152 ms | 0 - 423 MB | NPU
| Beit | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 36.169 ms | 0 - 3 MB | NPU
| Beit | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 6.695 ms | 0 - 3 MB | NPU
| Beit | ONNX | w8a16 | Qualcomm® QCS8450 | 11.152 ms | 0 - 423 MB | NPU
| Beit | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 6.624 ms | 0 - 3 MB | NPU
| Beit | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 6.899 ms | 96 - 96 MB | NPU
| Beit | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 3.578 ms | 0 - 362 MB | NPU
| Beit | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 3.578 ms | 0 - 362 MB | NPU
| Beit | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.398 ms | 0 - 257 MB | NPU
| Beit | QNN_DLC | float | Snapdragon® X2 Elite | 7.849 ms | 1 - 1 MB | NPU
| Beit | QNN_DLC | float | Snapdragon® X Elite | 15.125 ms | 1 - 1 MB | NPU
| Beit | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 10.498 ms | 0 - 392 MB | NPU
| Beit | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 17.579 ms | 0 - 380 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 47.915 ms | 1 - 295 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 14.39 ms | 1 - 415 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® SA8775P | 17.197 ms | 1 - 296 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® SA8650P | 17.197 ms | 1 - 296 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® SA8255P | 17.197 ms | 1 - 296 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® QCS8450 | 17.579 ms | 0 - 380 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 20.102 ms | 1 - 3 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 15.125 ms | 1 - 1 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 8.541 ms | 1 - 301 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® SA7255P | 47.915 ms | 1 - 295 MB | NPU
| Beit | QNN_DLC | float | Qualcomm® SA8295P | 15.339 ms | 1 - 289 MB | NPU
| Beit | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 8.541 ms | 1 - 301 MB | NPU
| Beit | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.181 ms | 1 - 301 MB | NPU
| Beit | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 3.156 ms | 0 - 0 MB | NPU
| Beit | QNN_DLC | w8a16 | Snapdragon® X Elite | 7.302 ms | 0 - 0 MB | NPU
| Beit | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 4.65 ms | 0 - 409 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 15.087 ms | 0 - 350 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 6.776 ms | 0 - 266 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® SA8775P | 7.041 ms | 0 - 351 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® SA8650P | 7.041 ms | 0 - 351 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® SA8255P | 7.041 ms | 0 - 351 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 6.843 ms | 0 - 2 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 7.302 ms | 0 - 0 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 65.256 ms | 0 - 428 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 8.652 ms | 0 - 400 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 3.557 ms | 0 - 351 MB | NPU
| Beit | QNN_DLC | w8a16 | Qualcomm® SA7255P | 15.087 ms | 0 - 350 MB | NPU
| Beit | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 3.557 ms | 0 - 351 MB | NPU
| Beit | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.554 ms | 0 - 244 MB | NPU
| Beit | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 8.652 ms | 0 - 400 MB | NPU
| Beit | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 10.452 ms | 0 - 404 MB | NPU
| Beit | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 17.698 ms | 0 - 371 MB | NPU
| Beit | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 47.98 ms | 0 - 300 MB | NPU
| Beit | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 14.039 ms | 0 - 2 MB | NPU
| Beit | TFLITE | float | Qualcomm® SA8775P | 17.172 ms | 0 - 302 MB | NPU
| Beit | TFLITE | float | Qualcomm® SA8650P | 17.172 ms | 0 - 302 MB | NPU
| Beit | TFLITE | float | Qualcomm® SA8255P | 17.172 ms | 0 - 302 MB | NPU
| Beit | TFLITE | float | Qualcomm® QCS8450 | 17.698 ms | 0 - 371 MB | NPU
| Beit | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 17.569 ms | 0 - 186 MB | NPU
| Beit | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 8.545 ms | 0 - 309 MB | NPU
| Beit | TFLITE | float | Qualcomm® SA7255P | 47.98 ms | 0 - 300 MB | NPU
| Beit | TFLITE | float | Qualcomm® SA8295P | 15.33 ms | 0 - 293 MB | NPU
| Beit | TFLITE | float | Snapdragon® 8 Elite Mobile | 8.545 ms | 0 - 309 MB | NPU
| Beit | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.181 ms | 0 - 312 MB | NPU

## License
* The license for the original implementation of Beit can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [BEIT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254)
* [Source Model Implementation](https://github.com/microsoft/unilm/tree/master/beit)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).