Hugging Face's logo

qualcomm
/
Distil-Whisper

Automatic Speech Recognition
PyTorch
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
5157d3f
·
verified ·
1 Parent(s): 7fe41cf

v0.50.2

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.50.2 for changelog.

Files changed (3) hide show
  1. LICENSE +1 -0
  2. README.md +133 -0
  3. release_assets.json +1 -0
LICENSE ADDED
@@ -0,0 +1 @@
 
 
1
+ The license of the original trained model can be found at https://github.com/huggingface/distil-whisper/blob/main/LICENSE.
README.md ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: other
4
+ tags:
5
+ - android
6
+ pipeline_tag: automatic-speech-recognition
7
+
8
+ ---
9
+
10
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/web-assets/model_demo.png)
11
+
12
+ # Distil-Whisper: Optimized for Qualcomm Devices
13
+
14
+ Distil-Whisper Small English is a distilled version of Whisper Small, optimized for fast and efficient automatic speech recognition.
15
+
16
+ This is based on the implementation of Distil-Whisper found [here](https://github.com/huggingface/distil-whisper).
17
+ This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/distil_whisper) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
18
+
19
+ Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
20
+
21
+ ## Getting Started
22
+ There are two ways to deploy this model on your device:
23
+
24
+ ### Option 1: Download Pre-Exported Models
25
+
26
+ Below are pre-exported model assets ready for deployment.
27
+
28
+ | Runtime | Precision | Chipset | SDK Versions | Download |
29
+ |---|---|---|---|---|
30
+ | ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.50.2/distil_whisper-onnx-float.zip)
31
+ | QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.50.2/distil_whisper-qnn_dlc-float.zip)
32
+ | TFLITE | float | Universal | QAIRT 2.43, TFLite 2.19.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.50.2/distil_whisper-tflite-float.zip)
33
+
34
+ For more device-specific assets and performance metrics, visit **[Distil-Whisper on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/distil_whisper)**.
35
+
36
+
37
+ ### Option 2: Export with Custom Configurations
38
+
39
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/distil_whisper) Python library to compile and export the model with your own:
40
+ - Custom weights (e.g., fine-tuned checkpoints)
41
+ - Custom input shapes
42
+ - Target device and runtime configurations
43
+
44
+ This option is ideal if you need to customize the model beyond the default configuration provided here.
45
+
46
+ See our repository for [Distil-Whisper on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/distil_whisper) for usage instructions.
47
+
48
+ ## Model Details
49
+
50
+ **Model Type:** Model_use_case.speech_recognition
51
+
52
+ **Model Stats:**
53
+ - Model checkpoint: distil-whisper/distil-small.en
54
+ - Input resolution: 80x3000 (30 seconds audio)
55
+ - Max decoded sequence length: 200 tokens
56
+ - Number of parameters (encoder): 166M
57
+ - Model size (encoder) (float): 332 MB
58
+ - Number of parameters (decoder): 211M
59
+ - Model size (decoder) (float): 450MB
60
+
61
+ ## Performance Summary
62
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
63
+ |---|---|---|---|---|---|---
64
+ | decoder | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.544 ms | 51 - 375 MB | NPU
65
+ | decoder | ONNX | float | Snapdragon® X2 Elite | 5.102 ms | 178 - 178 MB | NPU
66
+ | decoder | ONNX | float | Snapdragon® X Elite | 11.168 ms | 178 - 178 MB | NPU
67
+ | decoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.615 ms | 0 - 364 MB | NPU
68
+ | decoder | ONNX | float | Qualcomm® QCS8550 (Proxy) | 11.602 ms | 0 - 195 MB | NPU
69
+ | decoder | ONNX | float | Qualcomm® QCS9075 | 13.163 ms | 40 - 82 MB | NPU
70
+ | decoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.186 ms | 16 - 476 MB | NPU
71
+ | decoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.544 ms | 17 - 326 MB | NPU
72
+ | decoder | QNN_DLC | float | Snapdragon® X2 Elite | 5.615 ms | 40 - 40 MB | NPU
73
+ | decoder | QNN_DLC | float | Snapdragon® X Elite | 10.866 ms | 40 - 40 MB | NPU
74
+ | decoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 8.59 ms | 17 - 330 MB | NPU
75
+ | decoder | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 18.893 ms | 30 - 246 MB | NPU
76
+ | decoder | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 11.254 ms | 40 - 42 MB | NPU
77
+ | decoder | QNN_DLC | float | Qualcomm® SA8775P | 12.908 ms | 19 - 236 MB | NPU
78
+ | decoder | QNN_DLC | float | Qualcomm® QCS9075 | 13.082 ms | 40 - 86 MB | NPU
79
+ | decoder | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 18.076 ms | 36 - 345 MB | NPU
80
+ | decoder | QNN_DLC | float | Qualcomm® SA7255P | 18.893 ms | 30 - 246 MB | NPU
81
+ | decoder | QNN_DLC | float | Qualcomm® SA8295P | 14.217 ms | 20 - 271 MB | NPU
82
+ | decoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.215 ms | 5 - 447 MB | NPU
83
+ | decoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.627 ms | 4 - 480 MB | NPU
84
+ | decoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 8.471 ms | 4 - 556 MB | NPU
85
+ | decoder | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 18.822 ms | 5 - 450 MB | NPU
86
+ | decoder | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 11.489 ms | 5 - 7 MB | NPU
87
+ | decoder | TFLITE | float | Qualcomm® SA8775P | 12.895 ms | 5 - 450 MB | NPU
88
+ | decoder | TFLITE | float | Qualcomm® QCS9075 | 13.074 ms | 2 - 266 MB | NPU
89
+ | decoder | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 18.592 ms | 5 - 524 MB | NPU
90
+ | decoder | TFLITE | float | Qualcomm® SA7255P | 18.822 ms | 5 - 450 MB | NPU
91
+ | decoder | TFLITE | float | Qualcomm® SA8295P | 14.156 ms | 5 - 443 MB | NPU
92
+ | decoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 6.994 ms | 3 - 504 MB | NPU
93
+ | encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 50.657 ms | 77 - 836 MB | NPU
94
+ | encoder | ONNX | float | Snapdragon® X2 Elite | 50.82 ms | 183 - 183 MB | NPU
95
+ | encoder | ONNX | float | Snapdragon® X Elite | 124.045 ms | 182 - 182 MB | NPU
96
+ | encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 82.002 ms | 86 - 1245 MB | NPU
97
+ | encoder | ONNX | float | Qualcomm® QCS8550 (Proxy) | 118.492 ms | 0 - 197 MB | NPU
98
+ | encoder | ONNX | float | Qualcomm® QCS9075 | 150.326 ms | 79 - 83 MB | NPU
99
+ | encoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 60.397 ms | 81 - 774 MB | NPU
100
+ | encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 49.034 ms | 0 - 742 MB | NPU
101
+ | encoder | QNN_DLC | float | Snapdragon® X2 Elite | 48.393 ms | 1 - 1 MB | NPU
102
+ | encoder | QNN_DLC | float | Snapdragon® X Elite | 121.663 ms | 1 - 1 MB | NPU
103
+ | encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 82.41 ms | 0 - 1046 MB | NPU
104
+ | encoder | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 397.445 ms | 1 - 731 MB | NPU
105
+ | encoder | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 118.837 ms | 1 - 3 MB | NPU
106
+ | encoder | QNN_DLC | float | Qualcomm® SA8775P | 139.846 ms | 1 - 733 MB | NPU
107
+ | encoder | QNN_DLC | float | Qualcomm® QCS9075 | 152.427 ms | 1 - 39 MB | NPU
108
+ | encoder | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 276.295 ms | 1 - 917 MB | NPU
109
+ | encoder | QNN_DLC | float | Qualcomm® SA7255P | 397.445 ms | 1 - 731 MB | NPU
110
+ | encoder | QNN_DLC | float | Qualcomm® SA8295P | 207.259 ms | 1 - 668 MB | NPU
111
+ | encoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 60.835 ms | 1 - 665 MB | NPU
112
+ | encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 419.256 ms | 41 - 81 MB | GPU
113
+ | encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 479.748 ms | 0 - 144 MB | GPU
114
+ | encoder | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 3116.266 ms | 38 - 82 MB | GPU
115
+ | encoder | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 655.878 ms | 0 - 313 MB | GPU
116
+ | encoder | TFLITE | float | Qualcomm® SA8775P | 1326.077 ms | 27 - 71 MB | GPU
117
+ | encoder | TFLITE | float | Qualcomm® QCS9075 | 1270.615 ms | 0 - 40 MB | GPU
118
+ | encoder | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 843.956 ms | 39 - 193 MB | GPU
119
+ | encoder | TFLITE | float | Qualcomm® SA7255P | 3116.266 ms | 38 - 82 MB | GPU
120
+ | encoder | TFLITE | float | Qualcomm® SA8295P | 667.309 ms | 40 - 83 MB | GPU
121
+ | encoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 409.596 ms | 41 - 80 MB | GPU
122
+
123
+ ## License
124
+ * The license for the original implementation of Distil-Whisper can be found
125
+ [here](https://github.com/huggingface/distil-whisper/blob/main/LICENSE).
126
+
127
+ ## References
128
+ * [Distil-Whisper - Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://arxiv.org/abs/2311.00430)
129
+ * [Source Model Implementation](https://github.com/huggingface/distil-whisper)
130
+
131
+ ## Community
132
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
133
+ * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
release_assets.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"version":"0.50.2","precisions":{"float":{"universal_assets":{"onnx":{"tool_versions":{"qairt":"2.42.0.251225135753_193295","onnx_runtime":"1.24.3"},"download_url":"https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.50.2/distil_whisper-onnx-float.zip"},"qnn_dlc":{"tool_versions":{"qairt":"2.43.0.260127150333_193827"},"download_url":"https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.50.2/distil_whisper-qnn_dlc-float.zip"},"tflite":{"tool_versions":{"qairt":"2.43.0.260127150333_193827","tflite":"2.19.1"},"download_url":"https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.50.2/distil_whisper-tflite-float.zip"}}}}}