Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -36,52 +36,56 @@ More details on model performance across various devices, can be found
|
|
| 36 |
|
| 37 |
| Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
|
| 38 |
|---|---|---|---|---|---|---|---|---|
|
| 39 |
-
| TrOCREncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 50.
|
| 40 |
-
| TrOCREncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN |
|
| 41 |
-
| TrOCREncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 38.
|
| 42 |
-
| TrOCREncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE |
|
| 43 |
-
| TrOCREncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 41.
|
| 44 |
-
| TrOCREncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX |
|
| 45 |
-
| TrOCREncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE |
|
| 46 |
-
| TrOCREncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN |
|
| 47 |
-
| TrOCREncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 25.
|
| 48 |
-
| TrOCREncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 50.
|
| 49 |
-
| TrOCREncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 36.
|
| 50 |
-
| TrOCREncoder |
|
| 51 |
-
| TrOCREncoder |
|
| 52 |
-
| TrOCREncoder |
|
| 53 |
-
| TrOCREncoder |
|
| 54 |
-
| TrOCREncoder |
|
| 55 |
-
| TrOCREncoder |
|
| 56 |
-
| TrOCREncoder |
|
| 57 |
-
| TrOCREncoder |
|
| 58 |
-
| TrOCREncoder |
|
| 59 |
-
| TrOCREncoder |
|
| 60 |
-
| TrOCREncoder |
|
| 61 |
-
| TrOCREncoder |
|
| 62 |
-
|
|
| 63 |
-
|
|
| 64 |
-
| TrOCRDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 |
|
| 65 |
-
| TrOCRDecoder | Samsung Galaxy
|
| 66 |
-
| TrOCRDecoder | Samsung Galaxy
|
| 67 |
-
| TrOCRDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 |
|
| 68 |
-
| TrOCRDecoder |
|
| 69 |
-
| TrOCRDecoder |
|
| 70 |
-
| TrOCRDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite |
|
| 71 |
-
| TrOCRDecoder |
|
| 72 |
-
| TrOCRDecoder |
|
| 73 |
-
| TrOCRDecoder |
|
| 74 |
-
| TrOCRDecoder |
|
| 75 |
-
| TrOCRDecoder |
|
| 76 |
-
| TrOCRDecoder |
|
| 77 |
-
| TrOCRDecoder |
|
| 78 |
-
| TrOCRDecoder |
|
| 79 |
-
| TrOCRDecoder | SA8295P ADP | SA8295P | TFLITE | 3.
|
| 80 |
-
| TrOCRDecoder | SA8295P ADP | SA8295P | QNN | 3.
|
| 81 |
-
| TrOCRDecoder |
|
| 82 |
-
| TrOCRDecoder |
|
| 83 |
-
| TrOCRDecoder |
|
| 84 |
-
| TrOCRDecoder |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
|
| 86 |
|
| 87 |
|
|
@@ -146,8 +150,8 @@ Profiling Results
|
|
| 146 |
TrOCREncoder
|
| 147 |
Device : Samsung Galaxy S23 (13)
|
| 148 |
Runtime : TFLITE
|
| 149 |
-
Estimated inference time (ms) : 50.
|
| 150 |
-
Estimated peak memory usage (MB): [7,
|
| 151 |
Total # Ops : 591
|
| 152 |
Compute Unit(s) : NPU (591 ops)
|
| 153 |
|
|
@@ -155,8 +159,8 @@ Compute Unit(s) : NPU (591 ops)
|
|
| 155 |
TrOCRDecoder
|
| 156 |
Device : Samsung Galaxy S23 (13)
|
| 157 |
Runtime : TFLITE
|
| 158 |
-
Estimated inference time (ms) : 2.
|
| 159 |
-
Estimated peak memory usage (MB): [0,
|
| 160 |
Total # Ops : 399
|
| 161 |
Compute Unit(s) : NPU (399 ops)
|
| 162 |
```
|
|
@@ -177,11 +181,12 @@ in memory using the `jit.trace` and then call the `submit_compile_job` API.
|
|
| 177 |
import torch
|
| 178 |
|
| 179 |
import qai_hub as hub
|
| 180 |
-
from qai_hub_models.models.trocr import
|
| 181 |
|
| 182 |
# Load the model
|
| 183 |
-
|
| 184 |
-
|
|
|
|
| 185 |
|
| 186 |
# Device
|
| 187 |
device = hub.Device("Samsung Galaxy S23")
|
|
|
|
| 36 |
|
| 37 |
| Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
|
| 38 |
|---|---|---|---|---|---|---|---|---|
|
| 39 |
+
| TrOCREncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 50.056 ms | 7 - 31 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 40 |
+
| TrOCREncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 52.421 ms | 2 - 23 MB | FP16 | NPU | [TrOCR.so](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.so) |
|
| 41 |
+
| TrOCREncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 38.114 ms | 0 - 56 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.onnx) |
|
| 42 |
+
| TrOCREncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 39.56 ms | 30 - 89 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 43 |
+
| TrOCREncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 41.432 ms | 2 - 60 MB | FP16 | NPU | [TrOCR.so](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.so) |
|
| 44 |
+
| TrOCREncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 31.563 ms | 0 - 258 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.onnx) |
|
| 45 |
+
| TrOCREncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 35.481 ms | 7 - 69 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 46 |
+
| TrOCREncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 37.937 ms | 2 - 66 MB | FP16 | NPU | Use Export Script |
|
| 47 |
+
| TrOCREncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 25.867 ms | 9 - 129 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.onnx) |
|
| 48 |
+
| TrOCREncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 50.017 ms | 9 - 33 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 49 |
+
| TrOCREncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 36.736 ms | 2 - 3 MB | FP16 | NPU | Use Export Script |
|
| 50 |
+
| TrOCREncoder | SA7255P ADP | SA7255P | TFLITE | 266.256 ms | 7 - 69 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 51 |
+
| TrOCREncoder | SA7255P ADP | SA7255P | QNN | 248.746 ms | 2 - 9 MB | FP16 | NPU | Use Export Script |
|
| 52 |
+
| TrOCREncoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 50.949 ms | 7 - 31 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 53 |
+
| TrOCREncoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 37.437 ms | 2 - 10 MB | FP16 | NPU | Use Export Script |
|
| 54 |
+
| TrOCREncoder | SA8295P ADP | SA8295P | TFLITE | 65.315 ms | 7 - 67 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 55 |
+
| TrOCREncoder | SA8295P ADP | SA8295P | QNN | 50.613 ms | 2 - 8 MB | FP16 | NPU | Use Export Script |
|
| 56 |
+
| TrOCREncoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 50.607 ms | 7 - 34 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 57 |
+
| TrOCREncoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 37.746 ms | 2 - 3 MB | FP16 | NPU | Use Export Script |
|
| 58 |
+
| TrOCREncoder | SA8775P ADP | SA8775P | TFLITE | 59.808 ms | 6 - 67 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 59 |
+
| TrOCREncoder | SA8775P ADP | SA8775P | QNN | 42.298 ms | 0 - 5 MB | FP16 | NPU | Use Export Script |
|
| 60 |
+
| TrOCREncoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 60.447 ms | 7 - 70 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.tflite) |
|
| 61 |
+
| TrOCREncoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 63.93 ms | 0 - 61 MB | FP16 | NPU | Use Export Script |
|
| 62 |
+
| TrOCREncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 34.007 ms | 2 - 2 MB | FP16 | NPU | Use Export Script |
|
| 63 |
+
| TrOCREncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 36.795 ms | 49 - 49 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCREncoder.onnx) |
|
| 64 |
+
| TrOCRDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 2.21 ms | 0 - 323 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 65 |
+
| TrOCRDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 2.348 ms | 0 - 330 MB | FP16 | NPU | [TrOCR.so](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.so) |
|
| 66 |
+
| TrOCRDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 2.799 ms | 0 - 75 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.onnx) |
|
| 67 |
+
| TrOCRDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 1.694 ms | 8 - 57 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 68 |
+
| TrOCRDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 1.75 ms | 0 - 50 MB | FP16 | NPU | [TrOCR.so](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.so) |
|
| 69 |
+
| TrOCRDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 2.164 ms | 0 - 173 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.onnx) |
|
| 70 |
+
| TrOCRDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 1.573 ms | 0 - 44 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 71 |
+
| TrOCRDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 1.543 ms | 0 - 45 MB | FP16 | NPU | Use Export Script |
|
| 72 |
+
| TrOCRDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 2.079 ms | 0 - 132 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.onnx) |
|
| 73 |
+
| TrOCRDecoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 2.195 ms | 0 - 361 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 74 |
+
| TrOCRDecoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 2.249 ms | 1 - 3 MB | FP16 | NPU | Use Export Script |
|
| 75 |
+
| TrOCRDecoder | SA7255P ADP | SA7255P | TFLITE | 12.213 ms | 0 - 43 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 76 |
+
| TrOCRDecoder | SA7255P ADP | SA7255P | QNN | 12.403 ms | 7 - 12 MB | FP16 | NPU | Use Export Script |
|
| 77 |
+
| TrOCRDecoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 2.203 ms | 0 - 285 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 78 |
+
| TrOCRDecoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 2.373 ms | 2 - 3 MB | FP16 | NPU | Use Export Script |
|
| 79 |
+
| TrOCRDecoder | SA8295P ADP | SA8295P | TFLITE | 3.062 ms | 0 - 40 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 80 |
+
| TrOCRDecoder | SA8295P ADP | SA8295P | QNN | 3.494 ms | 7 - 13 MB | FP16 | NPU | Use Export Script |
|
| 81 |
+
| TrOCRDecoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 2.193 ms | 0 - 325 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 82 |
+
| TrOCRDecoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 2.268 ms | 1 - 3 MB | FP16 | NPU | Use Export Script |
|
| 83 |
+
| TrOCRDecoder | SA8775P ADP | SA8775P | TFLITE | 3.41 ms | 0 - 43 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 84 |
+
| TrOCRDecoder | SA8775P ADP | SA8775P | QNN | 3.581 ms | 7 - 13 MB | FP16 | NPU | Use Export Script |
|
| 85 |
+
| TrOCRDecoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 2.608 ms | 0 - 47 MB | FP16 | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.tflite) |
|
| 86 |
+
| TrOCRDecoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 2.863 ms | 6 - 54 MB | FP16 | NPU | Use Export Script |
|
| 87 |
+
| TrOCRDecoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 2.405 ms | 7 - 7 MB | FP16 | NPU | Use Export Script |
|
| 88 |
+
| TrOCRDecoder | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 2.74 ms | 68 - 68 MB | FP16 | NPU | [TrOCR.onnx](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCRDecoder.onnx) |
|
| 89 |
|
| 90 |
|
| 91 |
|
|
|
|
| 150 |
TrOCREncoder
|
| 151 |
Device : Samsung Galaxy S23 (13)
|
| 152 |
Runtime : TFLITE
|
| 153 |
+
Estimated inference time (ms) : 50.1
|
| 154 |
+
Estimated peak memory usage (MB): [7, 31]
|
| 155 |
Total # Ops : 591
|
| 156 |
Compute Unit(s) : NPU (591 ops)
|
| 157 |
|
|
|
|
| 159 |
TrOCRDecoder
|
| 160 |
Device : Samsung Galaxy S23 (13)
|
| 161 |
Runtime : TFLITE
|
| 162 |
+
Estimated inference time (ms) : 2.2
|
| 163 |
+
Estimated peak memory usage (MB): [0, 323]
|
| 164 |
Total # Ops : 399
|
| 165 |
Compute Unit(s) : NPU (399 ops)
|
| 166 |
```
|
|
|
|
| 181 |
import torch
|
| 182 |
|
| 183 |
import qai_hub as hub
|
| 184 |
+
from qai_hub_models.models.trocr import Model
|
| 185 |
|
| 186 |
# Load the model
|
| 187 |
+
model = Model.from_pretrained()
|
| 188 |
+
encoder_model = model.encoder
|
| 189 |
+
decoder_model = model.decoder
|
| 190 |
|
| 191 |
# Device
|
| 192 |
device = hub.Device("Samsung Galaxy S23")
|