Hugging Face's logo

qualcomm
/
Beit

Image Classification
PyTorch
backbone
android
Model card Files
xet
 Community
Beit
File size: 9,967 Bytes
8dac3ac
 
8b6a3af
8dac3ac
 
 
ab35384
8dac3ac
 
 
 
 
1b52625
8dac3ac
 
 
1b52625
081dcfe
1b52625
 
 
 
 
 
 
 
 
 
 
 
081dcfe
 
 
 
 
1b52625
 
 
 
 
 
081dcfe
1b52625
 
 
 
 
 
081dcfe
1b52625
 
 
 
 
 
 
 
 
 
 
 
 
 
081dcfe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8dac3ac
 
3568dd9
 
8dac3ac
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
---
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).