Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -38,44 +38,48 @@ More details on model performance across various devices, can be found
|
|
38 |
|
39 |
| Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
|
40 |
|---|---|---|---|---|---|---|---|---|
|
41 |
-
| WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE |
|
42 |
-
| WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN |
|
43 |
-
| WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE |
|
44 |
-
| WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN |
|
45 |
-
| WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE |
|
46 |
-
| WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN |
|
47 |
-
| WhisperEncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE |
|
48 |
-
| WhisperEncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN |
|
49 |
-
| WhisperEncoder |
|
50 |
-
| WhisperEncoder |
|
51 |
-
| WhisperEncoder |
|
52 |
-
| WhisperEncoder |
|
53 |
-
| WhisperEncoder |
|
54 |
-
| WhisperEncoder |
|
55 |
-
| WhisperEncoder |
|
56 |
-
| WhisperEncoder |
|
57 |
-
| WhisperEncoder |
|
58 |
-
| WhisperEncoder |
|
59 |
-
| WhisperEncoder |
|
60 |
-
|
|
61 |
-
|
|
62 |
-
| WhisperDecoder | Samsung Galaxy
|
63 |
-
| WhisperDecoder | Samsung Galaxy
|
64 |
-
| WhisperDecoder |
|
65 |
-
| WhisperDecoder |
|
66 |
-
| WhisperDecoder |
|
67 |
-
| WhisperDecoder |
|
68 |
-
| WhisperDecoder |
|
69 |
-
| WhisperDecoder |
|
70 |
-
| WhisperDecoder |
|
71 |
-
| WhisperDecoder |
|
72 |
-
| WhisperDecoder |
|
73 |
-
| WhisperDecoder |
|
74 |
| WhisperDecoder | SA8295P ADP | SA8295P | TFLITE | 4.786 ms | 3 - 29 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
75 |
-
| WhisperDecoder | SA8295P ADP | SA8295P | QNN | 3.
|
76 |
-
| WhisperDecoder |
|
77 |
-
| WhisperDecoder |
|
78 |
-
| WhisperDecoder |
|
|
|
|
|
|
|
|
|
79 |
|
80 |
|
81 |
|
@@ -140,8 +144,8 @@ Profiling Results
|
|
140 |
WhisperEncoder
|
141 |
Device : Samsung Galaxy S23 (13)
|
142 |
Runtime : TFLITE
|
143 |
-
Estimated inference time (ms) :
|
144 |
-
Estimated peak memory usage (MB): [15,
|
145 |
Total # Ops : 271
|
146 |
Compute Unit(s) : GPU (260 ops) CPU (11 ops)
|
147 |
|
@@ -149,8 +153,8 @@ Compute Unit(s) : GPU (260 ops) CPU (11 ops)
|
|
149 |
WhisperDecoder
|
150 |
Device : Samsung Galaxy S23 (13)
|
151 |
Runtime : TFLITE
|
152 |
-
Estimated inference time (ms) : 4.
|
153 |
-
Estimated peak memory usage (MB): [3,
|
154 |
Total # Ops : 557
|
155 |
Compute Unit(s) : NPU (557 ops)
|
156 |
```
|
@@ -171,11 +175,12 @@ in memory using the `jit.trace` and then call the `submit_compile_job` API.
|
|
171 |
import torch
|
172 |
|
173 |
import qai_hub as hub
|
174 |
-
from qai_hub_models.models.whisper_tiny_en import
|
175 |
|
176 |
# Load the model
|
177 |
-
|
178 |
-
|
|
|
179 |
|
180 |
# Device
|
181 |
device = hub.Device("Samsung Galaxy S23")
|
|
|
38 |
|
39 |
| Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
|
40 |
|---|---|---|---|---|---|---|---|---|
|
41 |
+
| WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 102.517 ms | 15 - 102 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
42 |
+
| WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 140.918 ms | 0 - 52 MB | FP16 | NPU | [Whisper-Tiny-En.so](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.so) |
|
43 |
+
| WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 79.931 ms | 20 - 49 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
44 |
+
| WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 110.457 ms | 0 - 190 MB | FP16 | NPU | [Whisper-Tiny-En.so](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.so) |
|
45 |
+
| WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 65.236 ms | 20 - 39 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
46 |
+
| WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 89.706 ms | 0 - 194 MB | FP16 | NPU | Use Export Script |
|
47 |
+
| WhisperEncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 100.71 ms | 20 - 109 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
48 |
+
| WhisperEncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 104.587 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
|
49 |
+
| WhisperEncoder | SA7255P ADP | SA7255P | TFLITE | 532.945 ms | 20 - 44 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
50 |
+
| WhisperEncoder | SA7255P ADP | SA7255P | QNN | 466.605 ms | 2 - 7 MB | FP16 | NPU | Use Export Script |
|
51 |
+
| WhisperEncoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 102.188 ms | 20 - 117 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
52 |
+
| WhisperEncoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 108.187 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
|
53 |
+
| WhisperEncoder | SA8295P ADP | SA8295P | TFLITE | 107.711 ms | 20 - 41 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
54 |
+
| WhisperEncoder | SA8295P ADP | SA8295P | QNN | 128.097 ms | 1 - 6 MB | FP16 | NPU | Use Export Script |
|
55 |
+
| WhisperEncoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 99.594 ms | 20 - 58 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
56 |
+
| WhisperEncoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 107.074 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
|
57 |
+
| WhisperEncoder | SA8775P ADP | SA8775P | TFLITE | 178.344 ms | 20 - 43 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
58 |
+
| WhisperEncoder | SA8775P ADP | SA8775P | QNN | 119.681 ms | 0 - 6 MB | FP16 | NPU | Use Export Script |
|
59 |
+
| WhisperEncoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 156.253 ms | 20 - 55 MB | FP16 | GPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperEncoder.tflite) |
|
60 |
+
| WhisperEncoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 158.505 ms | 0 - 194 MB | FP16 | NPU | Use Export Script |
|
61 |
+
| WhisperEncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 97.524 ms | 0 - 0 MB | FP16 | NPU | Use Export Script |
|
62 |
+
| WhisperDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 3.993 ms | 3 - 50 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
63 |
+
| WhisperDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 2.409 ms | 0 - 80 MB | FP16 | NPU | [Whisper-Tiny-En.so](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.so) |
|
64 |
+
| WhisperDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 3.077 ms | 0 - 58 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
65 |
+
| WhisperDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 1.726 ms | 0 - 42 MB | FP16 | NPU | [Whisper-Tiny-En.so](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.so) |
|
66 |
+
| WhisperDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 2.624 ms | 2 - 57 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
67 |
+
| WhisperDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 1.421 ms | 10 - 50 MB | FP16 | NPU | Use Export Script |
|
68 |
+
| WhisperDecoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 4.056 ms | 3 - 41 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
69 |
+
| WhisperDecoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 2.377 ms | 5 - 7 MB | FP16 | NPU | Use Export Script |
|
70 |
+
| WhisperDecoder | SA7255P ADP | SA7255P | TFLITE | 18.142 ms | 3 - 56 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
71 |
+
| WhisperDecoder | SA7255P ADP | SA7255P | QNN | 15.519 ms | 9 - 15 MB | FP16 | NPU | Use Export Script |
|
72 |
+
| WhisperDecoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 3.952 ms | 2 - 41 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
73 |
+
| WhisperDecoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 2.333 ms | 2 - 4 MB | FP16 | NPU | Use Export Script |
|
74 |
| WhisperDecoder | SA8295P ADP | SA8295P | TFLITE | 4.786 ms | 3 - 29 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
75 |
+
| WhisperDecoder | SA8295P ADP | SA8295P | QNN | 3.598 ms | 1 - 7 MB | FP16 | NPU | Use Export Script |
|
76 |
+
| WhisperDecoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 3.997 ms | 3 - 50 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
77 |
+
| WhisperDecoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 2.44 ms | 8 - 10 MB | FP16 | NPU | Use Export Script |
|
78 |
+
| WhisperDecoder | SA8775P ADP | SA8775P | TFLITE | 5.308 ms | 3 - 57 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
79 |
+
| WhisperDecoder | SA8775P ADP | SA8775P | QNN | 3.448 ms | 9 - 14 MB | FP16 | NPU | Use Export Script |
|
80 |
+
| WhisperDecoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 4.622 ms | 3 - 59 MB | FP16 | NPU | [Whisper-Tiny-En.tflite](https://huggingface.co/qualcomm/Whisper-Tiny-En/blob/main/WhisperDecoder.tflite) |
|
81 |
+
| WhisperDecoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 2.923 ms | 9 - 51 MB | FP16 | NPU | Use Export Script |
|
82 |
+
| WhisperDecoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 2.29 ms | 10 - 10 MB | FP16 | NPU | Use Export Script |
|
83 |
|
84 |
|
85 |
|
|
|
144 |
WhisperEncoder
|
145 |
Device : Samsung Galaxy S23 (13)
|
146 |
Runtime : TFLITE
|
147 |
+
Estimated inference time (ms) : 102.5
|
148 |
+
Estimated peak memory usage (MB): [15, 102]
|
149 |
Total # Ops : 271
|
150 |
Compute Unit(s) : GPU (260 ops) CPU (11 ops)
|
151 |
|
|
|
153 |
WhisperDecoder
|
154 |
Device : Samsung Galaxy S23 (13)
|
155 |
Runtime : TFLITE
|
156 |
+
Estimated inference time (ms) : 4.0
|
157 |
+
Estimated peak memory usage (MB): [3, 50]
|
158 |
Total # Ops : 557
|
159 |
Compute Unit(s) : NPU (557 ops)
|
160 |
```
|
|
|
175 |
import torch
|
176 |
|
177 |
import qai_hub as hub
|
178 |
+
from qai_hub_models.models.whisper_tiny_en import Model
|
179 |
|
180 |
# Load the model
|
181 |
+
model = Model.from_pretrained()
|
182 |
+
encoder_model = model.encoder
|
183 |
+
decoder_model = model.decoder
|
184 |
|
185 |
# Device
|
186 |
device = hub.Device("Samsung Galaxy S23")
|