Hugging Face's logo

qualcomm
/
VIT

Image Classification
PyTorch
LiteRT
backbone
android
Model card Files Community
qaihm-bot commited on
Commit
ac709b4
·
verified ·
1 Parent(s): a2072a1

v0.30.5

Browse files

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

Files changed (3) hide show
  1. README.md +68 -66
  2. VIT_w8a16.onnx +2 -2
  3. VIT_w8a8.onnx +2 -2
README.md CHANGED
@@ -35,68 +35,70 @@ 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
- | VIT | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 44.445 ms | 0 - 279 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
39
- | VIT | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 45.483 ms | 0 - 9 MB | NPU | Use Export Script |
40
- | VIT | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 18.787 ms | 0 - 288 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
41
- | VIT | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 21.241 ms | 1 - 303 MB | NPU | Use Export Script |
42
- | VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 13.173 ms | 0 - 23 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
43
- | VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 13.903 ms | 1 - 4 MB | NPU | Use Export Script |
44
- | VIT | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 16.14 ms | 0 - 279 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
45
- | VIT | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 16.84 ms | 1 - 10 MB | NPU | Use Export Script |
46
- | VIT | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 44.445 ms | 0 - 279 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
47
- | VIT | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 45.483 ms | 0 - 9 MB | NPU | Use Export Script |
48
- | VIT | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 12.775 ms | 0 - 12 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
49
- | VIT | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 13.943 ms | 1 - 3 MB | NPU | Use Export Script |
50
- | VIT | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 21.101 ms | 0 - 281 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
51
- | VIT | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 20.46 ms | 1 - 18 MB | NPU | Use Export Script |
52
- | VIT | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 12.8 ms | 0 - 23 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
53
- | VIT | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 13.947 ms | 1 - 2 MB | NPU | Use Export Script |
54
- | VIT | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 16.14 ms | 0 - 279 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
55
- | VIT | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 16.84 ms | 1 - 10 MB | NPU | Use Export Script |
56
- | VIT | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 13.19 ms | 0 - 22 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
57
- | VIT | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 13.823 ms | 0 - 32 MB | NPU | Use Export Script |
58
- | VIT | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 13.473 ms | 0 - 368 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
59
- | VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 9.052 ms | 3 - 287 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
60
- | VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 9.664 ms | 1 - 305 MB | NPU | Use Export Script |
61
- | VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 9.632 ms | 1 - 310 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
62
- | VIT | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 7.049 ms | 0 - 283 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
63
- | VIT | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 7.867 ms | 1 - 285 MB | NPU | Use Export Script |
64
- | VIT | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 6.753 ms | 1 - 287 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
65
- | VIT | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 14.729 ms | 1 - 1 MB | NPU | Use Export Script |
66
- | VIT | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 14.962 ms | 171 - 171 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
67
- | VIT | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 178.073 ms | 651 - 882 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a16.onnx) |
68
- | VIT | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 133.002 ms | 665 - 802 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a16.onnx) |
69
- | VIT | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 18.304 ms | 0 - 47 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
70
- | VIT | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 30.586 ms | 0 - 10 MB | NPU | Use Export Script |
71
- | VIT | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 14.222 ms | 0 - 55 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
72
- | VIT | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 14.697 ms | 0 - 160 MB | NPU | Use Export Script |
73
- | VIT | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 8.807 ms | 0 - 14 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
74
- | VIT | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 10.726 ms | 0 - 2 MB | NPU | Use Export Script |
75
- | VIT | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 9.219 ms | 0 - 47 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
76
- | VIT | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 9.704 ms | 0 - 11 MB | NPU | Use Export Script |
77
- | VIT | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 78.117 ms | 1 - 40 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
78
- | VIT | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN | 79.639 ms | 0 - 10 MB | NPU | Use Export Script |
79
- | VIT | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 18.304 ms | 0 - 47 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
80
- | VIT | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN | 30.586 ms | 0 - 10 MB | NPU | Use Export Script |
81
- | VIT | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 8.844 ms | 0 - 15 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
82
- | VIT | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 10.791 ms | 0 - 2 MB | NPU | Use Export Script |
83
- | VIT | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 16.007 ms | 0 - 50 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
84
- | VIT | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN | 16.824 ms | 0 - 17 MB | NPU | Use Export Script |
85
- | VIT | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 8.8 ms | 0 - 20 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
86
- | VIT | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 10.777 ms | 0 - 3 MB | NPU | Use Export Script |
87
- | VIT | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 9.219 ms | 0 - 47 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
88
- | VIT | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN | 9.704 ms | 0 - 11 MB | NPU | Use Export Script |
89
- | VIT | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 8.808 ms | 0 - 21 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
90
- | VIT | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 10.735 ms | 0 - 28 MB | NPU | Use Export Script |
91
- | VIT | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 25.254 ms | 0 - 118 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
92
- | VIT | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 6.322 ms | 0 - 52 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
93
- | VIT | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 7.157 ms | 0 - 151 MB | NPU | Use Export Script |
94
- | VIT | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 17.808 ms | 0 - 175 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
95
- | VIT | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 4.92 ms | 0 - 50 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
96
- | VIT | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 6.222 ms | 0 - 147 MB | NPU | Use Export Script |
97
- | VIT | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 14.851 ms | 0 - 170 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
98
- | VIT | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 11.359 ms | 0 - 0 MB | NPU | Use Export Script |
99
- | VIT | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 29.379 ms | 88 - 88 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
 
 
100
 
101
 
102
 
@@ -160,8 +162,8 @@ Profiling Results
160
  VIT
161
  Device : cs_8275 (ANDROID 14)
162
  Runtime : TFLITE
163
- Estimated inference time (ms) : 44.4
164
- Estimated peak memory usage (MB): [0, 279]
165
  Total # Ops : 1579
166
  Compute Unit(s) : npu (1579 ops) gpu (0 ops) cpu (0 ops)
167
  ```
@@ -250,13 +252,13 @@ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
250
  You can also run the demo on-device.
251
 
252
  ```bash
253
- python -m qai_hub_models.models.vit.demo --on-device
254
  ```
255
 
256
  **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
257
  environment, please add the following to your cell (instead of the above).
258
  ```
259
- %run -m qai_hub_models.models.vit.demo -- --on-device
260
  ```
261
 
262
 
 
35
 
36
  | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
+ | VIT | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 249.1 ms | 0 - 315 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
39
+ | VIT | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 248.995 ms | 1 - 10 MB | NPU | Use Export Script |
40
+ | VIT | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 17.91 ms | 0 - 317 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
41
+ | VIT | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 21.232 ms | 0 - 316 MB | NPU | Use Export Script |
42
+ | VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 13.083 ms | 0 - 15 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
43
+ | VIT | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 13.865 ms | 1 - 3 MB | NPU | Use Export Script |
44
+ | VIT | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 16.025 ms | 0 - 315 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
45
+ | VIT | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 16.89 ms | 1 - 10 MB | NPU | Use Export Script |
46
+ | VIT | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 249.1 ms | 0 - 315 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
47
+ | VIT | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 248.995 ms | 1 - 10 MB | NPU | Use Export Script |
48
+ | VIT | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 13.106 ms | 0 - 15 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
49
+ | VIT | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 13.886 ms | 1 - 4 MB | NPU | Use Export Script |
50
+ | VIT | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 20.029 ms | 0 - 307 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
51
+ | VIT | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 19.918 ms | 1 - 16 MB | NPU | Use Export Script |
52
+ | VIT | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 13.131 ms | 0 - 20 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
53
+ | VIT | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 13.943 ms | 1 - 4 MB | NPU | Use Export Script |
54
+ | VIT | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 16.025 ms | 0 - 315 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
55
+ | VIT | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 16.89 ms | 1 - 10 MB | NPU | Use Export Script |
56
+ | VIT | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 12.749 ms | 0 - 34 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
57
+ | VIT | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 13.849 ms | 0 - 28 MB | NPU | Use Export Script |
58
+ | VIT | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 13.585 ms | 0 - 293 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
59
+ | VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 9.013 ms | 0 - 323 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
60
+ | VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 9.579 ms | 1 - 333 MB | NPU | Use Export Script |
61
+ | VIT | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 9.249 ms | 138 - 474 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
62
+ | VIT | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 8.255 ms | 0 - 319 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT.tflite) |
63
+ | VIT | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 8.028 ms | 1 - 312 MB | NPU | Use Export Script |
64
+ | VIT | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 7.826 ms | 1 - 321 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
65
+ | VIT | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 14.611 ms | 1 - 1 MB | NPU | Use Export Script |
66
+ | VIT | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 14.889 ms | 171 - 171 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT.onnx) |
67
+ | VIT | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 154.348 ms | 652 - 904 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a16.onnx) |
68
+ | VIT | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 120.86 ms | 672 - 832 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a16.onnx) |
69
+ | VIT | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 116.224 ms | 678 - 809 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a16.onnx) |
70
+ | VIT | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 169.671 ms | 922 - 922 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a16.onnx) |
71
+ | VIT | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 53.751 ms | 0 - 51 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
72
+ | VIT | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 205.911 ms | 0 - 10 MB | NPU | Use Export Script |
73
+ | VIT | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 13.227 ms | 0 - 59 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
74
+ | VIT | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 14.52 ms | 0 - 201 MB | NPU | Use Export Script |
75
+ | VIT | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 12.431 ms | 0 - 44 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
76
+ | VIT | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 10.613 ms | 0 - 2 MB | NPU | Use Export Script |
77
+ | VIT | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 12.873 ms | 0 - 51 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
78
+ | VIT | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 9.671 ms | 0 - 10 MB | NPU | Use Export Script |
79
+ | VIT | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 51.37 ms | 2 - 45 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
80
+ | VIT | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN | 79.083 ms | 0 - 12 MB | NPU | Use Export Script |
81
+ | VIT | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 53.751 ms | 0 - 51 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
82
+ | VIT | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN | 205.911 ms | 0 - 10 MB | NPU | Use Export Script |
83
+ | VIT | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 12.564 ms | 0 - 21 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
84
+ | VIT | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 10.668 ms | 0 - 2 MB | NPU | Use Export Script |
85
+ | VIT | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 15.111 ms | 0 - 54 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
86
+ | VIT | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN | 16.868 ms | 0 - 17 MB | NPU | Use Export Script |
87
+ | VIT | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 12.595 ms | 0 - 19 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
88
+ | VIT | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 10.645 ms | 0 - 2 MB | NPU | Use Export Script |
89
+ | VIT | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 12.873 ms | 0 - 51 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
90
+ | VIT | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN | 9.671 ms | 0 - 10 MB | NPU | Use Export Script |
91
+ | VIT | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 12.493 ms | 0 - 20 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
92
+ | VIT | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 10.598 ms | 0 - 32 MB | NPU | Use Export Script |
93
+ | VIT | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 28.306 ms | 0 - 114 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
94
+ | VIT | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 8.944 ms | 0 - 56 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
95
+ | VIT | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 7.082 ms | 0 - 163 MB | NPU | Use Export Script |
96
+ | VIT | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 19.695 ms | 0 - 277 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
97
+ | VIT | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 6.133 ms | 0 - 58 MB | NPU | [VIT.tflite](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.tflite) |
98
+ | VIT | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 5.766 ms | 0 - 263 MB | NPU | Use Export Script |
99
+ | VIT | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 14.788 ms | 0 - 254 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
100
+ | VIT | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 11.218 ms | 1 - 1 MB | NPU | Use Export Script |
101
+ | VIT | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 34.257 ms | 88 - 88 MB | NPU | [VIT.onnx](https://huggingface.co/qualcomm/VIT/blob/main/VIT_w8a8.onnx) |
102
 
103
 
104
 
 
162
  VIT
163
  Device : cs_8275 (ANDROID 14)
164
  Runtime : TFLITE
165
+ Estimated inference time (ms) : 249.1
166
+ Estimated peak memory usage (MB): [0, 315]
167
  Total # Ops : 1579
168
  Compute Unit(s) : npu (1579 ops) gpu (0 ops) cpu (0 ops)
169
  ```
 
252
  You can also run the demo on-device.
253
 
254
  ```bash
255
+ python -m qai_hub_models.models.vit.demo --eval-mode on-device
256
  ```
257
 
258
  **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
259
  environment, please add the following to your cell (instead of the above).
260
  ```
261
+ %run -m qai_hub_models.models.vit.demo -- --eval-mode on-device
262
  ```
263
 
264
 
VIT_w8a16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:94aeca2df51a3828199b75d7b94fcbe999f85991c19049d2fe087c24ebf7b1a1
3
- size 347707307
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca1f62dd5209d2e9781b7328150ba225b195a0a1b46fbb420b08a7b754bd7772
3
+ size 347828616
VIT_w8a8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6a5ae4ed7defbcf8a87098f3927eb21dd025d3f83b57c88ddffccabb31d066bc
3
- size 347657458
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27990f99527adf4cf1acc8da76740f20cc045b7dd465b2fb68134fbad16fc658
3
+ size 347778767