Hugging Face's logo

qualcomm
/
ConvNext-Base

Image Classification
PyTorch
bu_auto
android
Model card Files
xet
Community
qaihm-bot commited on
Commit
e1a78fd
·
verified ·
1 Parent(s): efc149f

v0.46.1

Browse files

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

Files changed (7) hide show
  1. ConvNext-Base_float.dlc +0 -3
  2. ConvNext-Base_float.onnx.zip +0 -3
  3. ConvNext-Base_float.tflite +0 -3
  4. ConvNext-Base_w8a16.dlc +0 -3
  5. ConvNext-Base_w8a16.onnx.zip +0 -3
  6. README.md +88 -238
  7. tool-versions.yaml +0 -4
ConvNext-Base_float.dlc DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:62077d3acd5308bc2a6f5a49caa438379e9503dce5ba538dcd9e34c6c973ce27
3
- size 354737404
 
 
 
 
ConvNext-Base_float.onnx.zip DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:45e7449f548b053ec400cb1b8c6641c6881caa3e075f68c7e62006a901e7c31d
3
- size 329653263
 
 
 
 
ConvNext-Base_float.tflite DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:ca43f0a53cadef3ce71371d3a5cae8804202d51a9e031fd0a5324bccd3721804
3
- size 354419100
 
 
 
 
ConvNext-Base_w8a16.dlc DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:0716e993bab705bbde1d082592df9680b93bdcb2ead110d0d39eba849bffbd00
3
- size 93049620
 
 
 
 
ConvNext-Base_w8a16.onnx.zip DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:bf9c4396c93c4bce4a508ce47d769eb18c5678ca47e8ef6bdb5a17752a0f9178
3
- size 303882861
 
 
 
 
README.md CHANGED
@@ -10,256 +10,106 @@ pipeline_tag: image-classification
10
 
11
  ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/convnext_base/web-assets/model_demo.png)
12
 
13
- # ConvNext-Base: Optimized for Mobile Deployment
14
- ## Imagenet classifier and general purpose backbone
15
-
16
 
17
  ConvNextBase 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 ConvNext-Base found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py).
20
-
21
-
22
- This repository provides scripts to run ConvNext-Base on Qualcomm® devices.
23
- More details on model performance across various devices, can be found
24
- [here](https://aihub.qualcomm.com/models/convnext_base).
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: 88.6M
35
- - Model size (float): 338 MB
36
- - Model size (w8a16): 88.7 MB
37
-
38
- | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
39
- |---|---|---|---|---|---|---|---|---|
40
- | ConvNext-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 41.133 ms | 0 - 347 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
41
- | ConvNext-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 42.373 ms | 1 - 352 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
42
- | ConvNext-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 19.116 ms | 0 - 400 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
43
- | ConvNext-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 20.642 ms | 1 - 411 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
44
- | ConvNext-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 7.29 ms | 0 - 3 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
45
- | ConvNext-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 8.162 ms | 1 - 3 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
46
- | ConvNext-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 7.358 ms | 0 - 195 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.onnx.zip) |
47
- | ConvNext-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 11.123 ms | 0 - 346 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
48
- | ConvNext-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 11.985 ms | 1 - 352 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
49
- | ConvNext-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 5.534 ms | 0 - 414 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
50
- | ConvNext-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 6.075 ms | 1 - 421 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
51
- | ConvNext-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 5.445 ms | 0 - 394 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.onnx.zip) |
52
- | ConvNext-Base | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 4.128 ms | 0 - 345 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
53
- | ConvNext-Base | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 4.636 ms | 1 - 355 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
54
- | ConvNext-Base | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 4.223 ms | 0 - 327 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.onnx.zip) |
55
- | ConvNext-Base | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 3.188 ms | 0 - 349 MB | NPU | [ConvNext-Base.tflite](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.tflite) |
56
- | ConvNext-Base | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 3.548 ms | 1 - 355 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
57
- | ConvNext-Base | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 3.361 ms | 1 - 329 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.onnx.zip) |
58
- | ConvNext-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 8.551 ms | 1 - 1 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.dlc) |
59
- | ConvNext-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 7.436 ms | 176 - 176 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base.onnx.zip) |
60
- | ConvNext-Base | w8a16 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | QNN_DLC | 75.588 ms | 0 - 376 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
61
- | ConvNext-Base | w8a16 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | ONNX | 740.896 ms | 68 - 86 MB | CPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.onnx.zip) |
62
- | ConvNext-Base | w8a16 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | QNN_DLC | 23.777 ms | 0 - 2 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
63
- | ConvNext-Base | w8a16 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 1179.535 ms | 42 - 84 MB | CPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.onnx.zip) |
64
- | ConvNext-Base | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 14.483 ms | 0 - 282 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
65
- | ConvNext-Base | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 8.902 ms | 0 - 326 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
66
- | ConvNext-Base | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 5.883 ms | 0 - 3 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
67
- | ConvNext-Base | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 6.122 ms | 0 - 282 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
68
- | ConvNext-Base | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 4.181 ms | 0 - 332 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
69
- | ConvNext-Base | w8a16 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 3.263 ms | 0 - 276 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
70
- | ConvNext-Base | w8a16 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 249.621 ms | 55 - 203 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.onnx.zip) |
71
- | ConvNext-Base | w8a16 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 7.659 ms | 0 - 327 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
72
- | ConvNext-Base | w8a16 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 693.474 ms | 44 - 57 MB | CPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.onnx.zip) |
73
- | ConvNext-Base | w8a16 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 2.513 ms | 0 - 288 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
74
- | ConvNext-Base | w8a16 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 242.068 ms | 37 - 186 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.onnx.zip) |
75
- | ConvNext-Base | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 6.276 ms | 0 - 0 MB | NPU | [ConvNext-Base.dlc](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.dlc) |
76
- | ConvNext-Base | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 207.832 ms | 137 - 137 MB | NPU | [ConvNext-Base.onnx.zip](https://huggingface.co/qualcomm/ConvNext-Base/blob/main/ConvNext-Base_w8a16.onnx.zip) |
77
-
78
-
79
-
80
-
81
- ## Installation
82
-
83
-
84
- Install the package via pip:
85
- ```bash
86
- pip install qai-hub-models
87
- ```
88
-
89
-
90
- ## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device
91
-
92
- Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) with your
93
- Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`.
94
-
95
- With this API token, you can configure your client to run models on the cloud
96
- hosted devices.
97
- ```bash
98
- qai-hub configure --api_token API_TOKEN
99
- ```
100
- Navigate to [docs](https://workbench.aihub.qualcomm.com/docs/) for more information.
101
-
102
-
103
-
104
- ## Demo off target
105
-
106
- The package contains a simple end-to-end demo that downloads pre-trained
107
- weights and runs this model on a sample input.
108
-
109
- ```bash
110
- python -m qai_hub_models.models.convnext_base.demo
111
- ```
112
-
113
- The above demo runs a reference implementation of pre-processing, model
114
- inference, and post processing.
115
-
116
- **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
117
- environment, please add the following to your cell (instead of the above).
118
- ```
119
- %run -m qai_hub_models.models.convnext_base.demo
120
- ```
121
-
122
-
123
- ### Run model on a cloud-hosted device
124
-
125
- In addition to the demo, you can also run the model on a cloud-hosted Qualcomm®
126
- device. This script does the following:
127
- * Performance check on-device on a cloud-hosted device
128
- * Downloads compiled assets that can be deployed on-device for Android.
129
- * Accuracy check between PyTorch and on-device outputs.
130
-
131
- ```bash
132
- python -m qai_hub_models.models.convnext_base.export
133
- ```
134
-
135
-
136
-
137
- ## How does this work?
138
-
139
- This [export script](https://aihub.qualcomm.com/models/convnext_base/qai_hub_models/models/ConvNext-Base/export.py)
140
- leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
141
- on-device. Lets go through each step below in detail:
142
-
143
- Step 1: **Compile model for on-device deployment**
144
-
145
- To compile a PyTorch model for on-device deployment, we first trace the model
146
- in memory using the `jit.trace` and then call the `submit_compile_job` API.
147
-
148
- ```python
149
- import torch
150
-
151
- import qai_hub as hub
152
- from qai_hub_models.models.convnext_base import Model
153
-
154
- # Load the model
155
- torch_model = Model.from_pretrained()
156
-
157
- # Device
158
- device = hub.Device("Samsung Galaxy S25")
159
-
160
- # Trace model
161
- input_shape = torch_model.get_input_spec()
162
- sample_inputs = torch_model.sample_inputs()
163
-
164
- pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
165
-
166
- # Compile model on a specific device
167
- compile_job = hub.submit_compile_job(
168
- model=pt_model,
169
- device=device,
170
- input_specs=torch_model.get_input_spec(),
171
- )
172
-
173
- # Get target model to run on-device
174
- target_model = compile_job.get_target_model()
175
-
176
- ```
177
-
178
-
179
- Step 2: **Performance profiling on cloud-hosted device**
180
-
181
- After compiling models from step 1. Models can be profiled model on-device using the
182
- `target_model`. Note that this scripts runs the model on a device automatically
183
- provisioned in the cloud. Once the job is submitted, you can navigate to a
184
- provided job URL to view a variety of on-device performance metrics.
185
- ```python
186
- profile_job = hub.submit_profile_job(
187
- model=target_model,
188
- device=device,
189
- )
190
-
191
- ```
192
-
193
- Step 3: **Verify on-device accuracy**
194
-
195
- To verify the accuracy of the model on-device, you can run on-device inference
196
- on sample input data on the same cloud hosted device.
197
- ```python
198
- input_data = torch_model.sample_inputs()
199
- inference_job = hub.submit_inference_job(
200
- model=target_model,
201
- device=device,
202
- inputs=input_data,
203
- )
204
- on_device_output = inference_job.download_output_data()
205
-
206
- ```
207
- With the output of the model, you can compute like PSNR, relative errors or
208
- spot check the output with expected output.
209
-
210
- **Note**: This on-device profiling and inference requires access to Qualcomm®
211
- AI Hub Workbench. [Sign up for access](https://myaccount.qualcomm.com/signup).
212
-
213
-
214
-
215
- ## Run demo on a cloud-hosted device
216
-
217
- You can also run the demo on-device.
218
-
219
- ```bash
220
- python -m qai_hub_models.models.convnext_base.demo --eval-mode on-device
221
- ```
222
-
223
- **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
224
- environment, please add the following to your cell (instead of the above).
225
- ```
226
- %run -m qai_hub_models.models.convnext_base.demo -- --eval-mode on-device
227
- ```
228
-
229
-
230
- ## Deploying compiled model to Android
231
-
232
-
233
- The models can be deployed using multiple runtimes:
234
- - TensorFlow Lite (`.tflite` export): [This
235
- tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
236
- guide to deploy the .tflite model in an Android application.
237
-
238
-
239
- - QNN (`.so` export ): This [sample
240
- app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
241
- provides instructions on how to use the `.so` shared library in an Android application.
242
-
243
-
244
- ## View on Qualcomm® AI Hub
245
- Get more details on ConvNext-Base's performance across various devices [here](https://aihub.qualcomm.com/models/convnext_base).
246
- Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
247
-
248
 
249
  ## License
250
  * The license for the original implementation of ConvNext-Base can be found
251
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).
252
 
253
-
254
-
255
  ## References
256
  * [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545)
257
  * [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py)
258
 
259
-
260
-
261
  ## Community
262
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
263
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
264
-
265
-
 
10
 
11
  ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/convnext_base/web-assets/model_demo.png)
12
 
13
+ # ConvNext-Base: Optimized for Qualcomm Devices
 
 
14
 
15
  ConvNextBase 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 ConvNext-Base found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py).
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/convnext_base) 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/convnext_base/releases/v0.46.1/convnext_base-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/convnext_base/releases/v0.46.1/convnext_base-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/convnext_base/releases/v0.46.1/convnext_base-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/convnext_base/releases/v0.46.1/convnext_base-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/convnext_base/releases/v0.46.1/convnext_base-tflite-float.zip)
36
+
37
+ For more device-specific assets and performance metrics, visit **[ConvNext-Base on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/convnext_base)**.
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/convnext_base) 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 [ConvNext-Base on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/convnext_base) 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: 88.6M
59
+ - Model size (float): 338 MB
60
+ - Model size (w8a16): 88.7 MB
61
+
62
+ ## Performance Summary
63
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
64
+ |---|---|---|---|---|---|---
65
+ | ConvNext-Base | ONNX | float | Snapdragon® X Elite | 7.488 ms | 176 - 176 MB | NPU
66
+ | ConvNext-Base | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.436 ms | 0 - 394 MB | NPU
67
+ | ConvNext-Base | ONNX | float | Qualcomm® QCS8550 (Proxy) | 7.317 ms | 0 - 638 MB | NPU
68
+ | ConvNext-Base | ONNX | float | Qualcomm® QCS9075 | 11.598 ms | 0 - 4 MB | NPU
69
+ | ConvNext-Base | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.246 ms | 0 - 329 MB | NPU
70
+ | ConvNext-Base | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.352 ms | 0 - 332 MB | NPU
71
+ | ConvNext-Base | ONNX | w8a16 | Snapdragon® X Elite | 219.265 ms | 137 - 137 MB | NPU
72
+ | ConvNext-Base | ONNX | w8a16 | Qualcomm® QCS6490 | 1158.253 ms | 41 - 86 MB | CPU
73
+ | ConvNext-Base | ONNX | w8a16 | Qualcomm® QCS9075 | 317.703 ms | 93 - 96 MB | NPU
74
+ | ConvNext-Base | ONNX | w8a16 | Qualcomm® QCM6690 | 737.008 ms | 34 - 46 MB | CPU
75
+ | ConvNext-Base | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 268.327 ms | 78 - 229 MB | NPU
76
+ | ConvNext-Base | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 691.251 ms | 35 - 49 MB | CPU
77
+ | ConvNext-Base | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 237.073 ms | 89 - 240 MB | NPU
78
+ | ConvNext-Base | QNN_DLC | float | Snapdragon® X Elite | 8.589 ms | 1 - 1 MB | NPU
79
+ | ConvNext-Base | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 6.113 ms | 0 - 350 MB | NPU
80
+ | ConvNext-Base | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 42.453 ms | 1 - 279 MB | NPU
81
+ | ConvNext-Base | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 8.213 ms | 0 - 33 MB | NPU
82
+ | ConvNext-Base | QNN_DLC | float | Qualcomm® QCS9075 | 12.381 ms | 1 - 3 MB | NPU
83
+ | ConvNext-Base | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 20.603 ms | 0 - 337 MB | NPU
84
+ | ConvNext-Base | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.689 ms | 1 - 281 MB | NPU
85
+ | ConvNext-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.534 ms | 1 - 283 MB | NPU
86
+ | ConvNext-Base | QNN_DLC | w8a16 | Snapdragon® X Elite | 6.26 ms | 0 - 0 MB | NPU
87
+ | ConvNext-Base | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 4.106 ms | 0 - 248 MB | NPU
88
+ | ConvNext-Base | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 23.818 ms | 0 - 2 MB | NPU
89
+ | ConvNext-Base | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 14.472 ms | 0 - 199 MB | NPU
90
+ | ConvNext-Base | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 5.888 ms | 0 - 2 MB | NPU
91
+ | ConvNext-Base | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 6.122 ms | 0 - 2 MB | NPU
92
+ | ConvNext-Base | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 71.461 ms | 0 - 395 MB | NPU
93
+ | ConvNext-Base | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 9.182 ms | 0 - 246 MB | NPU
94
+ | ConvNext-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 3.31 ms | 0 - 190 MB | NPU
95
+ | ConvNext-Base | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 7.715 ms | 0 - 247 MB | NPU
96
+ | ConvNext-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.559 ms | 0 - 201 MB | NPU
97
+ | ConvNext-Base | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.533 ms | 0 - 345 MB | NPU
98
+ | ConvNext-Base | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 41.241 ms | 0 - 274 MB | NPU
99
+ | ConvNext-Base | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 7.334 ms | 0 - 3 MB | NPU
100
+ | ConvNext-Base | TFLITE | float | Qualcomm® QCS9075 | 11.149 ms | 0 - 177 MB | NPU
101
+ | ConvNext-Base | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 19.73 ms | 0 - 330 MB | NPU
102
+ | ConvNext-Base | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.167 ms | 0 - 277 MB | NPU
103
+ | ConvNext-Base | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.174 ms | 0 - 279 MB | NPU
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
104
 
105
  ## License
106
  * The license for the original implementation of ConvNext-Base can be found
107
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).
108
 
 
 
109
  ## References
110
  * [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545)
111
  * [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py)
112
 
 
 
113
  ## Community
114
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
115
  * 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