Hugging Face's logo

qualcomm
/
ConvNext-Tiny

Image Classification
PyTorch
LiteRT
ONNX
android
Model card Files Community
qaihm-bot commited on
Commit
7cc602b
·
verified ·
1 Parent(s): 82b51ba

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +40 -19
README.md CHANGED
@@ -17,7 +17,7 @@ tags:
17
 
18
  ConvNextTiny 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.
19
 
20
- This model is an implementation of ConvNext-Tiny found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py).
21
  This repository provides scripts to run ConvNext-Tiny on Qualcomm® devices.
22
  More details on model performance across various devices, can be found
23
  [here](https://aihub.qualcomm.com/models/convnext_tiny).
@@ -32,15 +32,32 @@ More details on model performance across various devices, can be found
32
  - Number of parameters: 28.6M
33
  - Model size: 109 MB
34
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
35
 
36
 
37
 
38
- | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
39
- | ---|---|---|---|---|---|---|---|
40
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | TFLite | 3.313 ms | 0 - 32 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite)
41
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | QNN Model Library | 3.839 ms | 0 - 130 MB | FP16 | NPU | [ConvNext-Tiny.so](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.so)
42
-
43
-
44
 
45
  ## Installation
46
 
@@ -95,16 +112,16 @@ device. This script does the following:
95
  ```bash
96
  python -m qai_hub_models.models.convnext_tiny.export
97
  ```
98
-
99
  ```
100
- Profile Job summary of ConvNext-Tiny
101
- --------------------------------------------------
102
- Device: Snapdragon X Elite CRD (11)
103
- Estimated Inference Time: 3.63 ms
104
- Estimated Peak Memory Range: 0.57-0.57 MB
105
- Compute Units: NPU (223) | Total (223)
106
-
107
-
 
108
  ```
109
 
110
 
@@ -203,15 +220,19 @@ provides instructions on how to use the `.so` shared library in an Android appl
203
  Get more details on ConvNext-Tiny's performance across various devices [here](https://aihub.qualcomm.com/models/convnext_tiny).
204
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
205
 
 
206
  ## License
207
- - The license for the original implementation of ConvNext-Tiny can be found
208
- [here](https://github.com/pytorch/vision/blob/main/LICENSE).
209
- - The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
 
210
 
211
  ## References
212
  * [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545)
213
  * [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py)
214
 
 
 
215
  ## Community
216
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
217
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
 
17
 
18
  ConvNextTiny 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.
19
 
20
+ This model is an implementation of ConvNext-Tiny found [here]({source_repo}).
21
  This repository provides scripts to run ConvNext-Tiny on Qualcomm® devices.
22
  More details on model performance across various devices, can be found
23
  [here](https://aihub.qualcomm.com/models/convnext_tiny).
 
32
  - Number of parameters: 28.6M
33
  - Model size: 109 MB
34
 
35
+ | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
36
+ |---|---|---|---|---|---|---|---|---|
37
+ | ConvNext-Tiny | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 3.402 ms | 0 - 3 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
38
+ | ConvNext-Tiny | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 3.892 ms | 1 - 89 MB | FP16 | NPU | [ConvNext-Tiny.so](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.so) |
39
+ | ConvNext-Tiny | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 13.414 ms | 1 - 4 MB | FP16 | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
40
+ | ConvNext-Tiny | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 2.577 ms | 0 - 208 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
41
+ | ConvNext-Tiny | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 3.299 ms | 1 - 35 MB | FP16 | NPU | [ConvNext-Tiny.so](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.so) |
42
+ | ConvNext-Tiny | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 9.798 ms | 0 - 372 MB | FP16 | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
43
+ | ConvNext-Tiny | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 3.342 ms | 0 - 2 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
44
+ | ConvNext-Tiny | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 3.633 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
45
+ | ConvNext-Tiny | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 3.385 ms | 1 - 3 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
46
+ | ConvNext-Tiny | SA8255 (Proxy) | SA8255P Proxy | QNN | 3.67 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
47
+ | ConvNext-Tiny | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 3.4 ms | 0 - 2 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
48
+ | ConvNext-Tiny | SA8775 (Proxy) | SA8775P Proxy | QNN | 3.67 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
49
+ | ConvNext-Tiny | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 3.384 ms | 0 - 2 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
50
+ | ConvNext-Tiny | SA8650 (Proxy) | SA8650P Proxy | QNN | 3.664 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
51
+ | ConvNext-Tiny | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 9.206 ms | 0 - 200 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
52
+ | ConvNext-Tiny | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 9.842 ms | 0 - 31 MB | FP16 | NPU | Use Export Script |
53
+ | ConvNext-Tiny | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 2.159 ms | 0 - 61 MB | FP16 | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
54
+ | ConvNext-Tiny | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 2.436 ms | 0 - 36 MB | FP16 | NPU | Use Export Script |
55
+ | ConvNext-Tiny | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 7.452 ms | 1 - 126 MB | FP16 | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
56
+ | ConvNext-Tiny | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 3.891 ms | 1 - 1 MB | FP16 | NPU | Use Export Script |
57
+ | ConvNext-Tiny | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 16.264 ms | 57 - 57 MB | FP16 | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
58
 
59
 
60
 
 
 
 
 
 
 
61
 
62
  ## Installation
63
 
 
112
  ```bash
113
  python -m qai_hub_models.models.convnext_tiny.export
114
  ```
 
115
  ```
116
+ Profiling Results
117
+ ------------------------------------------------------------
118
+ ConvNext-Tiny
119
+ Device : Samsung Galaxy S23 (13)
120
+ Runtime : TFLITE
121
+ Estimated inference time (ms) : 3.4
122
+ Estimated peak memory usage (MB): [0, 3]
123
+ Total # Ops : 328
124
+ Compute Unit(s) : NPU (328 ops)
125
  ```
126
 
127
 
 
220
  Get more details on ConvNext-Tiny's performance across various devices [here](https://aihub.qualcomm.com/models/convnext_tiny).
221
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
222
 
223
+
224
  ## License
225
+ * The license for the original implementation of ConvNext-Tiny can be found [here](https://github.com/pytorch/vision/blob/main/LICENSE).
226
+ * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
227
+
228
+
229
 
230
  ## References
231
  * [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545)
232
  * [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/convnext.py)
233
 
234
+
235
+
236
  ## Community
237
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
238
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).