qaihm-bot commited on
Commit
aa15323
·
verified ·
1 Parent(s): 8bba1d5

See https://github.com/quic/ai-hub-models/releases/v0.30.5 for changelog.

Files changed (3) hide show
  1. README.md +37 -36
  2. Swin-Base.onnx +2 -2
  3. Swin-Base_w8a16.onnx +2 -2
README.md CHANGED
@@ -35,38 +35,39 @@ More details on model performance across various devices, can be found
35
 
36
  | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
- | Swin-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 62.185 ms | 0 - 373 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
39
- | Swin-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 59.788 ms | 1 - 10 MB | NPU | Use Export Script |
40
- | Swin-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 29.441 ms | 0 - 377 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
41
- | Swin-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 39.241 ms | 0 - 454 MB | NPU | Use Export Script |
42
- | Swin-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 22.495 ms | 0 - 32 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
43
- | Swin-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 21.776 ms | 1 - 3 MB | NPU | Use Export Script |
44
- | Swin-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 25.865 ms | 0 - 373 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
45
- | Swin-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 24.687 ms | 1 - 10 MB | NPU | Use Export Script |
46
- | Swin-Base | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 62.185 ms | 0 - 373 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
47
- | Swin-Base | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 59.788 ms | 1 - 10 MB | NPU | Use Export Script |
48
- | Swin-Base | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 22.223 ms | 0 - 28 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
49
- | Swin-Base | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 21.555 ms | 1 - 3 MB | NPU | Use Export Script |
50
- | Swin-Base | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 32.09 ms | 0 - 361 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
51
- | Swin-Base | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 30.677 ms | 1 - 17 MB | NPU | Use Export Script |
52
- | Swin-Base | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 22.379 ms | 0 - 32 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
53
- | Swin-Base | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 21.809 ms | 1 - 3 MB | NPU | Use Export Script |
54
- | Swin-Base | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 25.865 ms | 0 - 373 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
55
- | Swin-Base | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 24.687 ms | 1 - 10 MB | NPU | Use Export Script |
56
- | Swin-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 22.423 ms | 0 - 31 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
57
- | Swin-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 21.662 ms | 0 - 41 MB | NPU | Use Export Script |
58
- | Swin-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 19.842 ms | 0 - 407 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
59
- | Swin-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 15.719 ms | 0 - 377 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
60
- | Swin-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 15.227 ms | 36 - 416 MB | NPU | Use Export Script |
61
- | Swin-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 13.725 ms | 1 - 375 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
62
- | Swin-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 14.523 ms | 0 - 375 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
63
- | Swin-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 12.548 ms | 1 - 332 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
64
- | Swin-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 22.036 ms | 1 - 1 MB | NPU | Use Export Script |
65
- | Swin-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 23.075 ms | 175 - 175 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
66
- | Swin-Base | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 164.594 ms | 525 - 812 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
67
- | Swin-Base | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 137.725 ms | 661 - 956 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
68
- | Swin-Base | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 125.603 ms | 614 - 879 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
69
- | Swin-Base | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 172.981 ms | 920 - 920 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
 
70
 
71
 
72
 
@@ -130,8 +131,8 @@ Profiling Results
130
  Swin-Base
131
  Device : cs_8275 (ANDROID 14)
132
  Runtime : TFLITE
133
- Estimated inference time (ms) : 62.2
134
- Estimated peak memory usage (MB): [0, 373]
135
  Total # Ops : 1568
136
  Compute Unit(s) : npu (1568 ops) gpu (0 ops) cpu (0 ops)
137
  ```
@@ -220,13 +221,13 @@ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
220
  You can also run the demo on-device.
221
 
222
  ```bash
223
- python -m qai_hub_models.models.swin_base.demo --on-device
224
  ```
225
 
226
  **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
227
  environment, please add the following to your cell (instead of the above).
228
  ```
229
- %run -m qai_hub_models.models.swin_base.demo -- --on-device
230
  ```
231
 
232
 
 
35
 
36
  | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
+ | Swin-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 296.467 ms | 0 - 386 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
39
+ | Swin-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 280.259 ms | 1 - 10 MB | NPU | Use Export Script |
40
+ | Swin-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 28.547 ms | 0 - 397 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
41
+ | Swin-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 28.388 ms | 1 - 378 MB | NPU | Use Export Script |
42
+ | Swin-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 22.272 ms | 0 - 36 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
43
+ | Swin-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 19.96 ms | 1 - 3 MB | NPU | Use Export Script |
44
+ | Swin-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 25.805 ms | 0 - 404 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
45
+ | Swin-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 22.934 ms | 0 - 10 MB | NPU | Use Export Script |
46
+ | Swin-Base | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 296.467 ms | 0 - 386 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
47
+ | Swin-Base | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 280.259 ms | 1 - 10 MB | NPU | Use Export Script |
48
+ | Swin-Base | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 22.078 ms | 0 - 40 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
49
+ | Swin-Base | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 19.827 ms | 1 - 3 MB | NPU | Use Export Script |
50
+ | Swin-Base | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 31.859 ms | 0 - 393 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
51
+ | Swin-Base | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 28.936 ms | 1 - 15 MB | NPU | Use Export Script |
52
+ | Swin-Base | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 22.282 ms | 0 - 44 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
53
+ | Swin-Base | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 20.025 ms | 1 - 3 MB | NPU | Use Export Script |
54
+ | Swin-Base | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 25.805 ms | 0 - 404 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
55
+ | Swin-Base | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 22.934 ms | 0 - 10 MB | NPU | Use Export Script |
56
+ | Swin-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 22.306 ms | 0 - 40 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
57
+ | Swin-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 20.104 ms | 0 - 65 MB | NPU | Use Export Script |
58
+ | Swin-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 19.632 ms | 0 - 387 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
59
+ | Swin-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 15.394 ms | 0 - 403 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
60
+ | Swin-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 13.543 ms | 1 - 387 MB | NPU | Use Export Script |
61
+ | Swin-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 13.214 ms | 1 - 391 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
62
+ | Swin-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 14.591 ms | 0 - 394 MB | NPU | [Swin-Base.tflite](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.tflite) |
63
+ | Swin-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 12.129 ms | 1 - 358 MB | NPU | Use Export Script |
64
+ | Swin-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 10.268 ms | 0 - 356 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
65
+ | Swin-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 20.862 ms | 1 - 1 MB | NPU | Use Export Script |
66
+ | Swin-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 22.288 ms | 175 - 175 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base.onnx) |
67
+ | Swin-Base | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 156.556 ms | 613 - 911 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
68
+ | Swin-Base | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 126.267 ms | 662 - 985 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
69
+ | Swin-Base | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 105.848 ms | 635 - 964 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
70
+ | Swin-Base | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 157.82 ms | 921 - 921 MB | NPU | [Swin-Base.onnx](https://huggingface.co/qualcomm/Swin-Base/blob/main/Swin-Base_w8a16.onnx) |
71
 
72
 
73
 
 
131
  Swin-Base
132
  Device : cs_8275 (ANDROID 14)
133
  Runtime : TFLITE
134
+ Estimated inference time (ms) : 296.5
135
+ Estimated peak memory usage (MB): [0, 386]
136
  Total # Ops : 1568
137
  Compute Unit(s) : npu (1568 ops) gpu (0 ops) cpu (0 ops)
138
  ```
 
221
  You can also run the demo on-device.
222
 
223
  ```bash
224
+ python -m qai_hub_models.models.swin_base.demo --eval-mode on-device
225
  ```
226
 
227
  **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
228
  environment, please add the following to your cell (instead of the above).
229
  ```
230
+ %run -m qai_hub_models.models.swin_base.demo -- --eval-mode on-device
231
  ```
232
 
233
 
Swin-Base.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4e4005e5a05ed4f231e83042df2c4baba0a60c56e0faba86bca965cfeadbde28
3
- size 355524848
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5abd0ab0ae6c72a96f75e3313f45c2261aed5cbed88bbd9d6b977e4ad59f8e66
3
+ size 355524911
Swin-Base_w8a16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c7f1dbf3bafe7367ea75006cc7c2d46f671ad74b0af4e6684a64f519767a914c
3
- size 356555812
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13ee8b8efa3f05d2d340b04115add1c5618208aa1f2cf11d8000b5e4a830d018
3
+ size 356637243