v0.30.5
Browse filesSee https://github.com/quic/ai-hub-models/releases/v0.30.5 for changelog.
- ConvNext-Tiny_w8a16.onnx +2 -2
- README.md +51 -45
ConvNext-Tiny_w8a16.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e10d51da836f96cdd505b21ec26e012b164d4789c259b20d8983812c54756ff
|
| 3 |
+
size 114746800
|
README.md
CHANGED
|
@@ -35,47 +35,53 @@ 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 |
-
| ConvNext-Tiny | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 14.
|
| 39 |
-
| ConvNext-Tiny | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN |
|
| 40 |
-
| ConvNext-Tiny | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE |
|
| 41 |
-
| ConvNext-Tiny | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN |
|
| 42 |
-
| ConvNext-Tiny | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE |
|
| 43 |
-
| ConvNext-Tiny | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 3.
|
| 44 |
-
| ConvNext-Tiny | float |
|
| 45 |
-
| ConvNext-Tiny | float |
|
| 46 |
-
| ConvNext-Tiny | float |
|
| 47 |
-
| ConvNext-Tiny | float |
|
| 48 |
-
| ConvNext-Tiny | float |
|
| 49 |
-
| ConvNext-Tiny | float |
|
| 50 |
-
| ConvNext-Tiny | float |
|
| 51 |
-
| ConvNext-Tiny | float |
|
| 52 |
-
| ConvNext-Tiny | float |
|
| 53 |
-
| ConvNext-Tiny | float |
|
| 54 |
-
| ConvNext-Tiny | float |
|
| 55 |
-
| ConvNext-Tiny | float |
|
| 56 |
-
| ConvNext-Tiny | float | Samsung Galaxy
|
| 57 |
-
| ConvNext-Tiny | float | Samsung Galaxy
|
| 58 |
-
| ConvNext-Tiny | float |
|
| 59 |
-
| ConvNext-Tiny | float |
|
| 60 |
-
| ConvNext-Tiny | float |
|
| 61 |
-
| ConvNext-Tiny | float |
|
| 62 |
-
| ConvNext-Tiny | float | Snapdragon
|
| 63 |
-
| ConvNext-Tiny |
|
| 64 |
-
| ConvNext-Tiny |
|
| 65 |
-
| ConvNext-Tiny |
|
| 66 |
-
| ConvNext-Tiny |
|
| 67 |
-
| ConvNext-Tiny | w8a16 |
|
| 68 |
-
| ConvNext-Tiny | w8a16 |
|
| 69 |
-
| ConvNext-Tiny | w8a16 |
|
| 70 |
-
| ConvNext-Tiny | w8a16 |
|
| 71 |
-
| ConvNext-Tiny | w8a16 |
|
| 72 |
-
| ConvNext-Tiny | w8a16 |
|
| 73 |
-
| ConvNext-Tiny | w8a16 |
|
| 74 |
-
| ConvNext-Tiny | w8a16 |
|
| 75 |
-
| ConvNext-Tiny | w8a16 |
|
| 76 |
-
| ConvNext-Tiny | w8a16 |
|
| 77 |
-
| ConvNext-Tiny | w8a16 |
|
| 78 |
-
| ConvNext-Tiny | w8a16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
|
| 80 |
|
| 81 |
|
|
@@ -139,8 +145,8 @@ Profiling Results
|
|
| 139 |
ConvNext-Tiny
|
| 140 |
Device : cs_8275 (ANDROID 14)
|
| 141 |
Runtime : TFLITE
|
| 142 |
-
Estimated inference time (ms) : 14.
|
| 143 |
-
Estimated peak memory usage (MB): [0,
|
| 144 |
Total # Ops : 328
|
| 145 |
Compute Unit(s) : npu (328 ops) gpu (0 ops) cpu (0 ops)
|
| 146 |
```
|
|
@@ -229,13 +235,13 @@ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
|
|
| 229 |
You can also run the demo on-device.
|
| 230 |
|
| 231 |
```bash
|
| 232 |
-
python -m qai_hub_models.models.convnext_tiny.demo --on-device
|
| 233 |
```
|
| 234 |
|
| 235 |
**NOTE**: If you want running in a Jupyter Notebook or Google Colab like
|
| 236 |
environment, please add the following to your cell (instead of the above).
|
| 237 |
```
|
| 238 |
-
%run -m qai_hub_models.models.convnext_tiny.demo -- --on-device
|
| 239 |
```
|
| 240 |
|
| 241 |
|
|
|
|
| 35 |
|
| 36 |
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
|
| 37 |
|---|---|---|---|---|---|---|---|---|
|
| 38 |
+
| ConvNext-Tiny | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 14.202 ms | 0 - 99 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 39 |
+
| ConvNext-Tiny | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 96.718 ms | 1 - 10 MB | NPU | Use Export Script |
|
| 40 |
+
| ConvNext-Tiny | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 7.824 ms | 0 - 108 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 41 |
+
| ConvNext-Tiny | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 9.996 ms | 1 - 110 MB | NPU | Use Export Script |
|
| 42 |
+
| ConvNext-Tiny | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 2.996 ms | 0 - 490 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 43 |
+
| ConvNext-Tiny | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 3.351 ms | 1 - 3 MB | NPU | Use Export Script |
|
| 44 |
+
| ConvNext-Tiny | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 4.244 ms | 0 - 99 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 45 |
+
| ConvNext-Tiny | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 4.729 ms | 1 - 11 MB | NPU | Use Export Script |
|
| 46 |
+
| ConvNext-Tiny | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 14.202 ms | 0 - 99 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 47 |
+
| ConvNext-Tiny | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 96.718 ms | 1 - 10 MB | NPU | Use Export Script |
|
| 48 |
+
| ConvNext-Tiny | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 2.992 ms | 0 - 490 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 49 |
+
| ConvNext-Tiny | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 3.384 ms | 1 - 3 MB | NPU | Use Export Script |
|
| 50 |
+
| ConvNext-Tiny | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 8.863 ms | 0 - 98 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 51 |
+
| ConvNext-Tiny | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 9.052 ms | 1 - 17 MB | NPU | Use Export Script |
|
| 52 |
+
| ConvNext-Tiny | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 2.947 ms | 0 - 476 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 53 |
+
| ConvNext-Tiny | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 3.352 ms | 1 - 3 MB | NPU | Use Export Script |
|
| 54 |
+
| ConvNext-Tiny | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 4.244 ms | 0 - 99 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 55 |
+
| ConvNext-Tiny | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 4.729 ms | 1 - 11 MB | NPU | Use Export Script |
|
| 56 |
+
| ConvNext-Tiny | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 2.991 ms | 0 - 489 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 57 |
+
| ConvNext-Tiny | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 3.339 ms | 1 - 13 MB | NPU | Use Export Script |
|
| 58 |
+
| ConvNext-Tiny | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 3.123 ms | 0 - 178 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
|
| 59 |
+
| ConvNext-Tiny | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 2.198 ms | 0 - 109 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 60 |
+
| ConvNext-Tiny | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 2.391 ms | 1 - 105 MB | NPU | Use Export Script |
|
| 61 |
+
| ConvNext-Tiny | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 2.199 ms | 0 - 109 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
|
| 62 |
+
| ConvNext-Tiny | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 1.948 ms | 0 - 100 MB | NPU | [ConvNext-Tiny.tflite](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.tflite) |
|
| 63 |
+
| ConvNext-Tiny | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 2.081 ms | 0 - 102 MB | NPU | Use Export Script |
|
| 64 |
+
| ConvNext-Tiny | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 1.997 ms | 1 - 102 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
|
| 65 |
+
| ConvNext-Tiny | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 3.592 ms | 1 - 1 MB | NPU | Use Export Script |
|
| 66 |
+
| ConvNext-Tiny | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 3.028 ms | 58 - 58 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny.onnx) |
|
| 67 |
+
| ConvNext-Tiny | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 6.311 ms | 0 - 10 MB | NPU | Use Export Script |
|
| 68 |
+
| ConvNext-Tiny | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 3.964 ms | 0 - 74 MB | NPU | Use Export Script |
|
| 69 |
+
| ConvNext-Tiny | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 2.845 ms | 0 - 3 MB | NPU | Use Export Script |
|
| 70 |
+
| ConvNext-Tiny | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 3.14 ms | 0 - 11 MB | NPU | Use Export Script |
|
| 71 |
+
| ConvNext-Tiny | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN | 11.874 ms | 0 - 13 MB | NPU | Use Export Script |
|
| 72 |
+
| ConvNext-Tiny | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN | 6.311 ms | 0 - 10 MB | NPU | Use Export Script |
|
| 73 |
+
| ConvNext-Tiny | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 2.845 ms | 0 - 2 MB | NPU | Use Export Script |
|
| 74 |
+
| ConvNext-Tiny | w8a16 | SA8295P ADP | Qualcomm® SA8295P | QNN | 4.283 ms | 0 - 18 MB | NPU | Use Export Script |
|
| 75 |
+
| ConvNext-Tiny | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 2.832 ms | 0 - 3 MB | NPU | Use Export Script |
|
| 76 |
+
| ConvNext-Tiny | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN | 3.14 ms | 0 - 11 MB | NPU | Use Export Script |
|
| 77 |
+
| ConvNext-Tiny | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 2.851 ms | 0 - 14 MB | NPU | Use Export Script |
|
| 78 |
+
| ConvNext-Tiny | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 87.074 ms | 210 - 359 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny_w8a16.onnx) |
|
| 79 |
+
| ConvNext-Tiny | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 1.971 ms | 0 - 68 MB | NPU | Use Export Script |
|
| 80 |
+
| ConvNext-Tiny | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 65.799 ms | 214 - 574 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny_w8a16.onnx) |
|
| 81 |
+
| ConvNext-Tiny | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 1.694 ms | 0 - 60 MB | NPU | Use Export Script |
|
| 82 |
+
| ConvNext-Tiny | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 53.12 ms | 213 - 543 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny_w8a16.onnx) |
|
| 83 |
+
| ConvNext-Tiny | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 3.109 ms | 0 - 0 MB | NPU | Use Export Script |
|
| 84 |
+
| ConvNext-Tiny | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 75.528 ms | 232 - 232 MB | NPU | [ConvNext-Tiny.onnx](https://huggingface.co/qualcomm/ConvNext-Tiny/blob/main/ConvNext-Tiny_w8a16.onnx) |
|
| 85 |
|
| 86 |
|
| 87 |
|
|
|
|
| 145 |
ConvNext-Tiny
|
| 146 |
Device : cs_8275 (ANDROID 14)
|
| 147 |
Runtime : TFLITE
|
| 148 |
+
Estimated inference time (ms) : 14.2
|
| 149 |
+
Estimated peak memory usage (MB): [0, 99]
|
| 150 |
Total # Ops : 328
|
| 151 |
Compute Unit(s) : npu (328 ops) gpu (0 ops) cpu (0 ops)
|
| 152 |
```
|
|
|
|
| 235 |
You can also run the demo on-device.
|
| 236 |
|
| 237 |
```bash
|
| 238 |
+
python -m qai_hub_models.models.convnext_tiny.demo --eval-mode on-device
|
| 239 |
```
|
| 240 |
|
| 241 |
**NOTE**: If you want running in a Jupyter Notebook or Google Colab like
|
| 242 |
environment, please add the following to your cell (instead of the above).
|
| 243 |
```
|
| 244 |
+
%run -m qai_hub_models.models.convnext_tiny.demo -- --eval-mode on-device
|
| 245 |
```
|
| 246 |
|
| 247 |
|