Hugging Face's logo

qualcomm
/
Swin-Base

Image Classification
PyTorch
TF Lite
ONNX
backbone
android
Model card Files Community
bhushans commited on
Commit
71318ee
·
verified ·
1 Parent(s): f1f314e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +44 -26
README.md CHANGED
@@ -35,29 +35,31 @@ More details on model performance across various devices, can be found
35
 
36
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
- | Swin-Base | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 25.772 ms | 0 - 4 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
39
- | Swin-Base | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 28.719 ms | 0 - 47 MB | FP16 | NPU | [Swin-Base.so](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.so) |
40
- | Swin-Base | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 46.518 ms | 0 - 226 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
41
- | Swin-Base | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 17.643 ms | 0 - 570 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
42
- | Swin-Base | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 20.02 ms | 0 - 198 MB | FP16 | NPU | [Swin-Base.so](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.so) |
43
- | Swin-Base | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 32.43 ms | 1 - 834 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
44
- | Swin-Base | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 16.394 ms | 0 - 268 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
45
- | Swin-Base | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 17.094 ms | 0 - 207 MB | FP16 | NPU | Use Export Script |
46
- | Swin-Base | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 29.188 ms | 1 - 330 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
47
- | Swin-Base | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 25.498 ms | 0 - 3 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
48
- | Swin-Base | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 26.715 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
49
- | Swin-Base | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 25.545 ms | 0 - 5 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
50
- | Swin-Base | SA8255 (Proxy) | SA8255P Proxy | QNN | 27.141 ms | 0 - 2 MB | FP16 | NPU | Use Export Script |
51
- | Swin-Base | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 25.447 ms | 0 - 3 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
52
- | Swin-Base | SA8775 (Proxy) | SA8775P Proxy | QNN | 27.434 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
53
- | Swin-Base | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 25.603 ms | 0 - 4 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
54
- | Swin-Base | SA8650 (Proxy) | SA8650P Proxy | QNN | 27.181 ms | 0 - 2 MB | FP16 | NPU | Use Export Script |
55
- | Swin-Base | SA8295P ADP | SA8295P | TFLITE | 36.059 ms | 0 - 242 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
56
- | Swin-Base | SA8295P ADP | SA8295P | QNN | 38.003 ms | 1 - 6 MB | FP16 | NPU | Use Export Script |
57
- | Swin-Base | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 32.521 ms | 0 - 539 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
58
- | Swin-Base | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 35.869 ms | 0 - 189 MB | FP16 | NPU | Use Export Script |
59
- | Swin-Base | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 27.678 ms | 1 - 1 MB | FP16 | NPU | Use Export Script |
60
- | Swin-Base | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 52.0 ms | 198 - 198 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
 
 
61
 
62
 
63
 
@@ -121,8 +123,8 @@ Profiling Results
121
  Swin-Base
122
  Device : Samsung Galaxy S23 (13)
123
  Runtime : TFLITE
124
- Estimated inference time (ms) : 25.8
125
- Estimated peak memory usage (MB): [0, 4]
126
  Total # Ops : 1568
127
  Compute Unit(s) : NPU (1568 ops)
128
  ```
@@ -143,13 +145,29 @@ in memory using the `jit.trace` and then call the `submit_compile_job` API.
143
  import torch
144
 
145
  import qai_hub as hub
146
- from qai_hub_models.models.swin_base import
147
 
148
  # Load the model
 
149
 
150
  # Device
151
  device = hub.Device("Samsung Galaxy S23")
152
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
153
 
154
  ```
155
 
 
35
 
36
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
+ | Swin-Base | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 26.392 ms | 0 - 40 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
39
+ | Swin-Base | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 26.202 ms | 0 - 46 MB | FP16 | NPU | [Swin-Base.so](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.so) |
40
+ | Swin-Base | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 43.136 ms | 0 - 194 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
41
+ | Swin-Base | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 18.329 ms | 0 - 201 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
42
+ | Swin-Base | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 17.993 ms | 0 - 201 MB | FP16 | NPU | [Swin-Base.so](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.so) |
43
+ | Swin-Base | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 29.943 ms | 1 - 802 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
44
+ | Swin-Base | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 16.682 ms | 0 - 202 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
45
+ | Swin-Base | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 16.586 ms | 0 - 208 MB | FP16 | NPU | Use Export Script |
46
+ | Swin-Base | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 27.271 ms | 0 - 300 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
47
+ | Swin-Base | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 26.291 ms | 0 - 43 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
48
+ | Swin-Base | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 23.602 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
49
+ | Swin-Base | SA7255P ADP | SA7255P | TFLITE | 307.272 ms | 0 - 205 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
50
+ | Swin-Base | SA7255P ADP | SA7255P | QNN | 303.06 ms | 1 - 6 MB | FP16 | NPU | Use Export Script |
51
+ | Swin-Base | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 26.172 ms | 0 - 45 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
52
+ | Swin-Base | SA8255 (Proxy) | SA8255P Proxy | QNN | 23.641 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
53
+ | Swin-Base | SA8295P ADP | SA8295P | TFLITE | 36.807 ms | 0 - 191 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
54
+ | Swin-Base | SA8295P ADP | SA8295P | QNN | 34.696 ms | 1 - 6 MB | FP16 | NPU | Use Export Script |
55
+ | Swin-Base | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 26.369 ms | 0 - 45 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
56
+ | Swin-Base | SA8650 (Proxy) | SA8650P Proxy | QNN | 23.763 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
57
+ | Swin-Base | SA8775P ADP | SA8775P | TFLITE | 35.687 ms | 0 - 206 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
58
+ | Swin-Base | SA8775P ADP | SA8775P | QNN | 33.079 ms | 0 - 5 MB | FP16 | NPU | Use Export Script |
59
+ | Swin-Base | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 33.515 ms | 0 - 192 MB | FP16 | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
60
+ | Swin-Base | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 33.69 ms | 1 - 189 MB | FP16 | NPU | Use Export Script |
61
+ | Swin-Base | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 24.377 ms | 1 - 1 MB | FP16 | NPU | Use Export Script |
62
+ | Swin-Base | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 50.182 ms | 175 - 175 MB | FP16 | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
63
 
64
 
65
 
 
123
  Swin-Base
124
  Device : Samsung Galaxy S23 (13)
125
  Runtime : TFLITE
126
+ Estimated inference time (ms) : 26.4
127
+ Estimated peak memory usage (MB): [0, 40]
128
  Total # Ops : 1568
129
  Compute Unit(s) : NPU (1568 ops)
130
  ```
 
145
  import torch
146
 
147
  import qai_hub as hub
148
+ from qai_hub_models.models.swin_base import Model
149
 
150
  # Load the model
151
+ torch_model = Model.from_pretrained()
152
 
153
  # Device
154
  device = hub.Device("Samsung Galaxy S23")
155
 
156
+ # Trace model
157
+ input_shape = torch_model.get_input_spec()
158
+ sample_inputs = torch_model.sample_inputs()
159
+
160
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
161
+
162
+ # Compile model on a specific device
163
+ compile_job = hub.submit_compile_job(
164
+ model=pt_model,
165
+ device=device,
166
+ input_specs=torch_model.get_input_spec(),
167
+ )
168
+
169
+ # Get target model to run on-device
170
+ target_model = compile_job.get_target_model()
171
 
172
  ```
173