Hugging Face's logo

qualcomm
/
Whisper-Small

Automatic Speech Recognition
PyTorch
foundation
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
ee2e0a1
·
verified ·
1 Parent(s): a15dc67

v0.49.1

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.49.1 for changelog.

Files changed (1) hide show
  1. README.md +59 -59
README.md CHANGED
@@ -15,7 +15,7 @@ pipeline_tag: automatic-speech-recognition
15
  HuggingFace Whisper-Small ASR (Automatic Speech Recognition) model is a state-of-the-art system designed for transcribing spoken language into written text. This model is based on the transformer architecture and has been optimized for edge inference by replacing Multi-Head Attention (MHA) with Single-Head Attention (SHA) and linear layers with convolutional (conv) layers. It exhibits robust performance in realistic, noisy environments, making it highly reliable for real-world applications. Specifically, it excels in long-form transcription, capable of accurately transcribing audio clips up to 30 seconds long. Time to the first token is the encoder's latency, while time to each additional token is decoder's latency, where we assume a max decoded length specified below.
16
 
17
  This is based on the implementation of Whisper-Small found [here](https://github.com/huggingface/transformers/tree/v4.42.3/src/transformers/models/whisper).
18
- 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/qai_hub_models/models/whisper_small) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
19
 
20
  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.
21
 
@@ -28,38 +28,38 @@ Below are pre-exported model assets ready for deployment.
28
 
29
  | Runtime | Precision | Chipset | SDK Versions | Download |
30
  |---|---|---|---|---|
31
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_x2_elite.zip)
32
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_x_elite.zip)
33
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_8gen3.zip)
34
- | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_qcs8550_proxy.zip)
35
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
36
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_gen5.zip)
37
- | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-precompiled_qnn_onnx-float-qualcomm_qcs9075.zip)
38
- | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_x2_elite.zip)
39
- | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_x_elite.zip)
40
- | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_8gen3.zip)
41
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_qcs8550_proxy.zip)
42
- | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_sa8775p.zip)
43
- | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
44
- | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_8_elite_gen5.zip)
45
- | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_sa7255p.zip)
46
- | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_sa8295p.zip)
47
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_qcs9075.zip)
48
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.48.0/whisper_small-qnn_context_binary-float-qualcomm_qcs8450_proxy.zip)
49
 
50
  For more device-specific assets and performance metrics, visit **[Whisper-Small on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/whisper_small)**.
51
 
52
 
53
  ### Option 2: Export with Custom Configurations
54
 
55
- Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/whisper_small) Python library to compile and export the model with your own:
56
  - Custom weights (e.g., fine-tuned checkpoints)
57
  - Custom input shapes
58
  - Target device and runtime configurations
59
 
60
  This option is ideal if you need to customize the model beyond the default configuration provided here.
61
 
62
- See our repository for [Whisper-Small on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/whisper_small) for usage instructions.
63
 
64
  ## Model Details
65
 
@@ -77,44 +77,44 @@ See our repository for [Whisper-Small on GitHub](https://github.com/qualcomm/ai-
77
  ## Performance Summary
78
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
79
  |---|---|---|---|---|---|---
80
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 6.066 ms | 286 - 286 MB | NPU
81
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 10.659 ms | 285 - 285 MB | NPU
82
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.924 ms | 75 - 82 MB | NPU
83
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 12.13 ms | 51 - 52 MB | NPU
84
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 13.199 ms | 60 - 122 MB | NPU
85
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.372 ms | 34 - 44 MB | NPU
86
- | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.22 ms | 75 - 85 MB | NPU
87
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 6.73 ms | 60 - 60 MB | NPU
88
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 10.092 ms | 60 - 60 MB | NPU
89
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 9.665 ms | 39 - 47 MB | NPU
90
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 19.565 ms | 49 - 56 MB | NPU
91
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 11.771 ms | 51 - 53 MB | NPU
92
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 32.478 ms | 48 - 56 MB | NPU
93
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 13.442 ms | 60 - 128 MB | NPU
94
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 17.843 ms | 60 - 70 MB | NPU
95
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 19.565 ms | 49 - 56 MB | NPU
96
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 14.883 ms | 49 - 55 MB | NPU
97
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.301 ms | 6 - 20 MB | NPU
98
- | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.263 ms | 60 - 71 MB | NPU
99
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 47.508 ms | 202 - 202 MB | NPU
100
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 115.666 ms | 201 - 201 MB | NPU
101
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 82.335 ms | 126 - 138 MB | NPU
102
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 113.587 ms | 0 - 225 MB | NPU
103
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 136.416 ms | 130 - 133 MB | NPU
104
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 59.579 ms | 128 - 140 MB | NPU
105
- | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 45.208 ms | 88 - 98 MB | NPU
106
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 47.808 ms | 0 - 0 MB | NPU
107
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 114.74 ms | 0 - 0 MB | NPU
108
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 85.474 ms | 1 - 9 MB | NPU
109
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 406.948 ms | 0 - 7 MB | NPU
110
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 112.603 ms | 0 - 3 MB | NPU
111
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 136.378 ms | 0 - 8 MB | NPU
112
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 136.967 ms | 0 - 55 MB | NPU
113
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 278.305 ms | 1 - 11 MB | NPU
114
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 406.948 ms | 0 - 7 MB | NPU
115
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 205.763 ms | 1 - 6 MB | NPU
116
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 60.203 ms | 1 - 10 MB | NPU
117
- | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 45.098 ms | 1 - 11 MB | NPU
118
 
119
  ## License
120
  * The license for the original implementation of Whisper-Small can be found
 
15
  HuggingFace Whisper-Small ASR (Automatic Speech Recognition) model is a state-of-the-art system designed for transcribing spoken language into written text. This model is based on the transformer architecture and has been optimized for edge inference by replacing Multi-Head Attention (MHA) with Single-Head Attention (SHA) and linear layers with convolutional (conv) layers. It exhibits robust performance in realistic, noisy environments, making it highly reliable for real-world applications. Specifically, it excels in long-form transcription, capable of accurately transcribing audio clips up to 30 seconds long. Time to the first token is the encoder's latency, while time to each additional token is decoder's latency, where we assume a max decoded length specified below.
16
 
17
  This is based on the implementation of Whisper-Small found [here](https://github.com/huggingface/transformers/tree/v4.42.3/src/transformers/models/whisper).
18
+ 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/tree/v0.49.1/qai_hub_models/models/whisper_small) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
19
 
20
  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.
21
 
 
28
 
29
  | Runtime | Precision | Chipset | SDK Versions | Download |
30
  |---|---|---|---|---|
31
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_gen5.zip)
32
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_x2_elite.zip)
33
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_x_elite.zip)
34
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_8gen3.zip)
35
+ | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_qcs8550_proxy.zip)
36
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
37
+ | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-precompiled_qnn_onnx-float-qualcomm_qcs9075.zip)
38
+ | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_8_elite_gen5.zip)
39
+ | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_x2_elite.zip)
40
+ | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_x_elite.zip)
41
+ | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_8gen3.zip)
42
+ | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_qcs8550_proxy.zip)
43
+ | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_sa8775p.zip)
44
+ | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
45
+ | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_sa7255p.zip)
46
+ | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_sa8295p.zip)
47
+ | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_qcs9075.zip)
48
+ | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/releases/v0.49.1/whisper_small-qnn_context_binary-float-qualcomm_qcs8450_proxy.zip)
49
 
50
  For more device-specific assets and performance metrics, visit **[Whisper-Small on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/whisper_small)**.
51
 
52
 
53
  ### Option 2: Export with Custom Configurations
54
 
55
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/tree/v0.49.1/qai_hub_models/models/whisper_small) Python library to compile and export the model with your own:
56
  - Custom weights (e.g., fine-tuned checkpoints)
57
  - Custom input shapes
58
  - Target device and runtime configurations
59
 
60
  This option is ideal if you need to customize the model beyond the default configuration provided here.
61
 
62
+ See our repository for [Whisper-Small on GitHub](https://github.com/qualcomm/ai-hub-models/tree/v0.49.1/qai_hub_models/models/whisper_small) for usage instructions.
63
 
64
  ## Model Details
65
 
 
77
  ## Performance Summary
78
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
79
  |---|---|---|---|---|---|---
80
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.223 ms | 71 - 81 MB | NPU
81
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 6.059 ms | 286 - 286 MB | NPU
82
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 10.763 ms | 285 - 285 MB | NPU
83
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.89 ms | 73 - 85 MB | NPU
84
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 12.424 ms | 61 - 63 MB | NPU
85
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 13.41 ms | 60 - 122 MB | NPU
86
+ | HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.313 ms | 16 - 27 MB | NPU
87
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.286 ms | 60 - 70 MB | NPU
88
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 6.573 ms | 60 - 60 MB | NPU
89
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 10.029 ms | 60 - 60 MB | NPU
90
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 9.935 ms | 37 - 45 MB | NPU
91
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 19.621 ms | 50 - 58 MB | NPU
92
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 12.084 ms | 38 - 40 MB | NPU
93
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 13.582 ms | 58 - 68 MB | NPU
94
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 13.447 ms | 60 - 128 MB | NPU
95
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 17.572 ms | 60 - 70 MB | NPU
96
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 19.621 ms | 50 - 58 MB | NPU
97
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 14.951 ms | 54 - 59 MB | NPU
98
+ | HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.35 ms | 0 - 10 MB | NPU
99
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 45.44 ms | 89 - 99 MB | NPU
100
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 47.338 ms | 202 - 202 MB | NPU
101
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 115.448 ms | 201 - 201 MB | NPU
102
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 82.504 ms | 128 - 135 MB | NPU
103
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 114.015 ms | 1 - 226 MB | NPU
104
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 137.592 ms | 126 - 129 MB | NPU
105
+ | HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 59.273 ms | 128 - 134 MB | NPU
106
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 45.32 ms | 1 - 11 MB | NPU
107
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 47.464 ms | 0 - 0 MB | NPU
108
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 114.871 ms | 0 - 0 MB | NPU
109
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 83.951 ms | 1 - 8 MB | NPU
110
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 405.778 ms | 0 - 8 MB | NPU
111
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 112.825 ms | 0 - 3 MB | NPU
112
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 136.523 ms | 0 - 9 MB | NPU
113
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 137.257 ms | 0 - 55 MB | NPU
114
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 280.808 ms | 1 - 10 MB | NPU
115
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 405.778 ms | 0 - 8 MB | NPU
116
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 205.718 ms | 0 - 5 MB | NPU
117
+ | HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 60.051 ms | 1 - 14 MB | NPU
118
 
119
  ## License
120
  * The license for the original implementation of Whisper-Small can be found