Hugging Face's logo

qualcomm
/
Mobile-VIT

Image Classification
PyTorch
backbone
android
Model card Files
xet
Community
qaihm-bot commited on
Commit
008f1af
·
verified ·
1 Parent(s): 912bff2

v0.46.1

Browse files

See https://github.com/quic/ai-hub-models/releases/v0.46.1 for changelog.

Files changed (6) hide show
  1. Mobile-VIT_float.dlc +0 -3
  2. Mobile-VIT_float.onnx.zip +0 -3
  3. Mobile-VIT_float.tflite +0 -3
  4. Mobile-VIT_w8a16.onnx.zip +0 -3
  5. README.md +85 -236
  6. tool-versions.yaml +0 -4
Mobile-VIT_float.dlc DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:64215ec44874a32e3514e6809dc0d423b4fbc4d0edf0151d022bdc35864cf4ce
3
- size 22728852
 
 
 
 
Mobile-VIT_float.onnx.zip DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:187509fe11e18ba1e53640109d3334bccf113ad212f53a975ef0156f60dad4fa
3
- size 20771378
 
 
 
 
Mobile-VIT_float.tflite DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:b483759032d6f469cc5663dd2497e0bd7f4a6728e340f90d722b1abdc638a1d6
3
- size 22400344
 
 
 
 
Mobile-VIT_w8a16.onnx.zip DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:fc3c1de8d9d9d3e3e94bbadcfdda393023b98976194da9f7660e1a3eb2908eb1
3
- size 14824089
 
 
 
 
README.md CHANGED
@@ -10,254 +10,103 @@ pipeline_tag: image-classification
10
 
11
  ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/web-assets/model_demo.png)
12
 
13
- # Mobile-VIT: Optimized for Mobile Deployment
14
- ## Imagenet classifier and general purpose backbone
15
-
16
 
17
  MobileVit 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.
18
 
19
- This model is an implementation of Mobile-VIT found [here](https://github.com/apple/ml-cvnets).
20
-
21
-
22
- This repository provides scripts to run Mobile-VIT on Qualcomm® devices.
23
- More details on model performance across various devices, can be found
24
- [here](https://aihub.qualcomm.com/models/mobile_vit).
25
-
26
-
27
-
28
- ### Model Details
29
-
30
- - **Model Type:** Model_use_case.image_classification
31
- - **Model Stats:**
32
- - Model checkpoint: Imagenet
33
- - Input resolution: 224x224
34
- - Number of parameters: 5.57M
35
- - Model size (float): 21.4 MB
36
- - Model size (w8a16): 6.56 MB
37
-
38
- | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
39
- |---|---|---|---|---|---|---|---|---|
40
- | Mobile-VIT | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 10.138 ms | 0 - 156 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
41
- | Mobile-VIT | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 9.88 ms | 1 - 155 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
42
- | Mobile-VIT | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 6.006 ms | 0 - 186 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
43
- | Mobile-VIT | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 5.726 ms | 1 - 186 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
44
- | Mobile-VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 3.625 ms | 0 - 3 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
45
- | Mobile-VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.475 ms | 1 - 3 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
46
- | Mobile-VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 4.581 ms | 0 - 16 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.onnx.zip) |
47
- | Mobile-VIT | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 4.455 ms | 0 - 147 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
48
- | Mobile-VIT | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 18.723 ms | 1 - 156 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
49
- | Mobile-VIT | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 10.138 ms | 0 - 156 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
50
- | Mobile-VIT | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 9.88 ms | 1 - 155 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
51
- | Mobile-VIT | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 6.637 ms | 0 - 165 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
52
- | Mobile-VIT | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 6.473 ms | 1 - 161 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
53
- | Mobile-VIT | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 4.455 ms | 0 - 147 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
54
- | Mobile-VIT | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 18.723 ms | 1 - 156 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
55
- | Mobile-VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 2.543 ms | 0 - 189 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
56
- | Mobile-VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.436 ms | 1 - 188 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
57
- | Mobile-VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.116 ms | 0 - 172 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.onnx.zip) |
58
- | Mobile-VIT | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 2.005 ms | 0 - 151 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
59
- | Mobile-VIT | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 1.906 ms | 1 - 160 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
60
- | Mobile-VIT | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 2.531 ms | 0 - 138 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.onnx.zip) |
61
- | Mobile-VIT | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 1.65 ms | 0 - 158 MB | NPU | [Mobile-VIT.tflite](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.tflite) |
62
- | Mobile-VIT | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 1.614 ms | 1 - 149 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
63
- | Mobile-VIT | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 2.14 ms | 0 - 139 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.onnx.zip) |
64
- | Mobile-VIT | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 3.819 ms | 1 - 1 MB | NPU | [Mobile-VIT.dlc](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.dlc) |
65
- | Mobile-VIT | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 4.686 ms | 12 - 12 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT.onnx.zip) |
66
- | Mobile-VIT | w8a16 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | ONNX | 148.326 ms | 66 - 83 MB | CPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
67
- | Mobile-VIT | w8a16 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 334.849 ms | 63 - 67 MB | CPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
68
- | Mobile-VIT | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 19.439 ms | 7 - 18 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
69
- | Mobile-VIT | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 15.2 ms | 12 - 188 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
70
- | Mobile-VIT | w8a16 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 12.51 ms | 13 - 151 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
71
- | Mobile-VIT | w8a16 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 132.5 ms | 63 - 81 MB | CPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
72
- | Mobile-VIT | w8a16 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 11.779 ms | 12 - 151 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
73
- | Mobile-VIT | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 16.823 ms | 16 - 16 MB | NPU | [Mobile-VIT.onnx.zip](https://huggingface.co/qualcomm/Mobile-VIT/blob/main/Mobile-VIT_w8a16.onnx.zip) |
74
-
75
-
76
-
77
-
78
- ## Installation
79
-
80
-
81
- Install the package via pip:
82
- ```bash
83
- # NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
84
- pip install "qai-hub-models[mobile-vit]"
85
- ```
86
-
87
-
88
- ## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device
89
-
90
- Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) with your
91
- Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`.
92
-
93
- With this API token, you can configure your client to run models on the cloud
94
- hosted devices.
95
- ```bash
96
- qai-hub configure --api_token API_TOKEN
97
- ```
98
- Navigate to [docs](https://workbench.aihub.qualcomm.com/docs/) for more information.
99
-
100
-
101
-
102
- ## Demo off target
103
-
104
- The package contains a simple end-to-end demo that downloads pre-trained
105
- weights and runs this model on a sample input.
106
-
107
- ```bash
108
- python -m qai_hub_models.models.mobile_vit.demo
109
- ```
110
-
111
- The above demo runs a reference implementation of pre-processing, model
112
- inference, and post processing.
113
-
114
- **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
115
- environment, please add the following to your cell (instead of the above).
116
- ```
117
- %run -m qai_hub_models.models.mobile_vit.demo
118
- ```
119
-
120
-
121
- ### Run model on a cloud-hosted device
122
-
123
- In addition to the demo, you can also run the model on a cloud-hosted Qualcomm®
124
- device. This script does the following:
125
- * Performance check on-device on a cloud-hosted device
126
- * Downloads compiled assets that can be deployed on-device for Android.
127
- * Accuracy check between PyTorch and on-device outputs.
128
-
129
- ```bash
130
- python -m qai_hub_models.models.mobile_vit.export
131
- ```
132
-
133
-
134
-
135
- ## How does this work?
136
-
137
- This [export script](https://aihub.qualcomm.com/models/mobile_vit/qai_hub_models/models/Mobile-VIT/export.py)
138
- leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
139
- on-device. Lets go through each step below in detail:
140
-
141
- Step 1: **Compile model for on-device deployment**
142
-
143
- To compile a PyTorch model for on-device deployment, we first trace the model
144
- in memory using the `jit.trace` and then call the `submit_compile_job` API.
145
-
146
- ```python
147
- import torch
148
-
149
- import qai_hub as hub
150
- from qai_hub_models.models.mobile_vit import Model
151
-
152
- # Load the model
153
- torch_model = Model.from_pretrained()
154
-
155
- # Device
156
- device = hub.Device("Samsung Galaxy S25")
157
-
158
- # Trace model
159
- input_shape = torch_model.get_input_spec()
160
- sample_inputs = torch_model.sample_inputs()
161
-
162
- pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
163
-
164
- # Compile model on a specific device
165
- compile_job = hub.submit_compile_job(
166
- model=pt_model,
167
- device=device,
168
- input_specs=torch_model.get_input_spec(),
169
- )
170
-
171
- # Get target model to run on-device
172
- target_model = compile_job.get_target_model()
173
-
174
- ```
175
-
176
-
177
- Step 2: **Performance profiling on cloud-hosted device**
178
-
179
- After compiling models from step 1. Models can be profiled model on-device using the
180
- `target_model`. Note that this scripts runs the model on a device automatically
181
- provisioned in the cloud. Once the job is submitted, you can navigate to a
182
- provided job URL to view a variety of on-device performance metrics.
183
- ```python
184
- profile_job = hub.submit_profile_job(
185
- model=target_model,
186
- device=device,
187
- )
188
-
189
- ```
190
-
191
- Step 3: **Verify on-device accuracy**
192
-
193
- To verify the accuracy of the model on-device, you can run on-device inference
194
- on sample input data on the same cloud hosted device.
195
- ```python
196
- input_data = torch_model.sample_inputs()
197
- inference_job = hub.submit_inference_job(
198
- model=target_model,
199
- device=device,
200
- inputs=input_data,
201
- )
202
- on_device_output = inference_job.download_output_data()
203
-
204
- ```
205
- With the output of the model, you can compute like PSNR, relative errors or
206
- spot check the output with expected output.
207
-
208
- **Note**: This on-device profiling and inference requires access to Qualcomm®
209
- AI Hub Workbench. [Sign up for access](https://myaccount.qualcomm.com/signup).
210
-
211
-
212
-
213
- ## Run demo on a cloud-hosted device
214
-
215
- You can also run the demo on-device.
216
-
217
- ```bash
218
- python -m qai_hub_models.models.mobile_vit.demo --eval-mode on-device
219
- ```
220
-
221
- **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
222
- environment, please add the following to your cell (instead of the above).
223
- ```
224
- %run -m qai_hub_models.models.mobile_vit.demo -- --eval-mode on-device
225
- ```
226
-
227
-
228
- ## Deploying compiled model to Android
229
-
230
-
231
- The models can be deployed using multiple runtimes:
232
- - TensorFlow Lite (`.tflite` export): [This
233
- tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
234
- guide to deploy the .tflite model in an Android application.
235
-
236
-
237
- - QNN (`.so` export ): This [sample
238
- app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
239
- provides instructions on how to use the `.so` shared library in an Android application.
240
-
241
-
242
- ## View on Qualcomm® AI Hub
243
- Get more details on Mobile-VIT's performance across various devices [here](https://aihub.qualcomm.com/models/mobile_vit).
244
- Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
245
-
246
 
247
  ## License
248
  * The license for the original implementation of Mobile-VIT can be found
249
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).
250
 
251
-
252
-
253
  ## References
254
  * [MOBILEVIT: LIGHT-WEIGHT, GENERAL-PURPOSE, AND MOBILE-FRIENDLY VISION TRANSFORMER](https://arxiv.org/abs/2110.02178)
255
  * [Source Model Implementation](https://github.com/apple/ml-cvnets)
256
 
257
-
258
-
259
  ## Community
260
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
261
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
262
-
263
-
 
10
 
11
  ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/web-assets/model_demo.png)
12
 
13
+ # Mobile-VIT: Optimized for Qualcomm Devices
 
 
14
 
15
  MobileVit 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.
16
 
17
+ This is based on the implementation of Mobile-VIT found [here](https://github.com/apple/ml-cvnets).
18
+ This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobile_vit) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
19
+
20
+ 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.
21
+
22
+ ## Getting Started
23
+ There are two ways to deploy this model on your device:
24
+
25
+ ### Option 1: Download Pre-Exported Models
26
+
27
+ Below are pre-exported model assets ready for deployment.
28
+
29
+ | Runtime | Precision | Chipset | SDK Versions | Download |
30
+ |---|---|---|---|---|
31
+ | ONNX | float | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/releases/v0.46.1/mobile_vit-onnx-float.zip)
32
+ | ONNX | w8a16 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/releases/v0.46.1/mobile_vit-onnx-w8a16.zip)
33
+ | QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/releases/v0.46.1/mobile_vit-qnn_dlc-float.zip)
34
+ | QNN_DLC | w8a16 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/releases/v0.46.1/mobile_vit-qnn_dlc-w8a16.zip)
35
+ | TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobile_vit/releases/v0.46.1/mobile_vit-tflite-float.zip)
36
+
37
+ For more device-specific assets and performance metrics, visit **[Mobile-VIT on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/mobile_vit)**.
38
+
39
+
40
+ ### Option 2: Export with Custom Configurations
41
+
42
+ Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobile_vit) Python library to compile and export the model with your own:
43
+ - Custom weights (e.g., fine-tuned checkpoints)
44
+ - Custom input shapes
45
+ - Target device and runtime configurations
46
+
47
+ This option is ideal if you need to customize the model beyond the default configuration provided here.
48
+
49
+ See our repository for [Mobile-VIT on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobile_vit) for usage instructions.
50
+
51
+ ## Model Details
52
+
53
+ **Model Type:** Model_use_case.image_classification
54
+
55
+ **Model Stats:**
56
+ - Model checkpoint: Imagenet
57
+ - Input resolution: 224x224
58
+ - Number of parameters: 5.57M
59
+ - Model size (float): 21.4 MB
60
+ - Model size (w8a16): 6.56 MB
61
+
62
+ ## Performance Summary
63
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
64
+ |---|---|---|---|---|---|---
65
+ | Mobile-VIT | ONNX | float | Snapdragon® X Elite | 4.702 ms | 12 - 12 MB | NPU
66
+ | Mobile-VIT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 3.142 ms | 0 - 168 MB | NPU
67
+ | Mobile-VIT | ONNX | float | Qualcomm® QCS8550 (Proxy) | 4.544 ms | 0 - 133 MB | NPU
68
+ | Mobile-VIT | ONNX | float | Qualcomm® QCS9075 | 5.619 ms | 1 - 4 MB | NPU
69
+ | Mobile-VIT | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.534 ms | 0 - 139 MB | NPU
70
+ | Mobile-VIT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.145 ms | 0 - 139 MB | NPU
71
+ | Mobile-VIT | ONNX | w8a16 | Snapdragon® X Elite | 16.897 ms | 16 - 16 MB | NPU
72
+ | Mobile-VIT | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 15.32 ms | 13 - 189 MB | NPU
73
+ | Mobile-VIT | ONNX | w8a16 | Qualcomm® QCS6490 | 339.344 ms | 69 - 73 MB | CPU
74
+ | Mobile-VIT | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 18.861 ms | 7 - 17 MB | NPU
75
+ | Mobile-VIT | ONNX | w8a16 | Qualcomm® QCS9075 | 20.873 ms | 12 - 14 MB | NPU
76
+ | Mobile-VIT | ONNX | w8a16 | Qualcomm® QCM6690 | 149.372 ms | 63 - 74 MB | CPU
77
+ | Mobile-VIT | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 12.17 ms | 13 - 152 MB | NPU
78
+ | Mobile-VIT | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 131.756 ms | 63 - 73 MB | CPU
79
+ | Mobile-VIT | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 11.754 ms | 12 - 151 MB | NPU
80
+ | Mobile-VIT | QNN_DLC | float | Snapdragon® X Elite | 3.849 ms | 1 - 1 MB | NPU
81
+ | Mobile-VIT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.456 ms | 0 - 98 MB | NPU
82
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 9.874 ms | 1 - 59 MB | NPU
83
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 3.48 ms | 0 - 13 MB | NPU
84
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® SA8775P | 4.259 ms | 1 - 61 MB | NPU
85
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® QCS9075 | 4.47 ms | 1 - 3 MB | NPU
86
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 5.699 ms | 0 - 94 MB | NPU
87
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® SA7255P | 9.874 ms | 1 - 59 MB | NPU
88
+ | Mobile-VIT | QNN_DLC | float | Qualcomm® SA8295P | 6.44 ms | 1 - 63 MB | NPU
89
+ | Mobile-VIT | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.907 ms | 1 - 68 MB | NPU
90
+ | Mobile-VIT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.611 ms | 1 - 73 MB | NPU
91
+ | Mobile-VIT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.58 ms | 0 - 108 MB | NPU
92
+ | Mobile-VIT | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 10.21 ms | 0 - 85 MB | NPU
93
+ | Mobile-VIT | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 3.652 ms | 0 - 19 MB | NPU
94
+ | Mobile-VIT | TFLITE | float | Qualcomm® SA8775P | 4.461 ms | 0 - 80 MB | NPU
95
+ | Mobile-VIT | TFLITE | float | Qualcomm® QCS9075 | 4.555 ms | 0 - 15 MB | NPU
96
+ | Mobile-VIT | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 6.017 ms | 0 - 94 MB | NPU
97
+ | Mobile-VIT | TFLITE | float | Qualcomm® SA7255P | 10.21 ms | 0 - 85 MB | NPU
98
+ | Mobile-VIT | TFLITE | float | Qualcomm® SA8295P | 6.691 ms | 0 - 71 MB | NPU
99
+ | Mobile-VIT | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.016 ms | 0 - 74 MB | NPU
100
+ | Mobile-VIT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.644 ms | 0 - 84 MB | NPU
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
101
 
102
  ## License
103
  * The license for the original implementation of Mobile-VIT can be found
104
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).
105
 
 
 
106
  ## References
107
  * [MOBILEVIT: LIGHT-WEIGHT, GENERAL-PURPOSE, AND MOBILE-FRIENDLY VISION TRANSFORMER](https://arxiv.org/abs/2110.02178)
108
  * [Source Model Implementation](https://github.com/apple/ml-cvnets)
109
 
 
 
110
  ## Community
111
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
112
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
 
 
tool-versions.yaml DELETED
@@ -1,4 +0,0 @@
1
- tool_versions:
2
- onnx:
3
- qairt: 2.37.1.250807093845_124904
4
- onnx_runtime: 1.23.0