| --- |
| license: apache-2.0 |
| pipeline_tag: automatic-speech-recognition |
| library_name: executorch |
| --- |
| |
| # whisper-tiny |
|
|
| This repository hosts the **whisper-tiny** models exported for the |
| [React Native ExecuTorch](https://www.npmjs.com/package/react-native-executorch) |
| library as ExecuTorch `.pte` programs, ready to run on device. |
|
|
| Upstream model: [whisper-tiny](https://huggingface.co/openai/whisper-tiny) |
|
|
| ## Variants |
|
|
| | Path | Backend | Precision | |
| | --- | --- | --- | |
| | `coreml/whisper_tiny_coreml_fp16.pte` | coreml | fp16 | |
| | `mlx/whisper_tiny_mlx_bf16.pte` | mlx | bf16 | |
| | `mlx/whisper_tiny_mlx_int8.pte` | mlx | int8 | |
| | `vulkan/whisper_tiny_vulkan_fp16.pte` | vulkan | fp16 | |
| | `vulkan/whisper_tiny_vulkan_int8.pte` | vulkan | int8 | |
| | `xnnpack/whisper_tiny_xnnpack_fp32.pte` | xnnpack | fp32 | |
| | `xnnpack/whisper_tiny_xnnpack_int8.pte` | xnnpack | int8 | |
|
|
| ## XNNPACK int8 |
|
|
| `xnnpack/whisper_tiny_xnnpack_int8.pte` is 8da8w (int8 dynamic activations, int8 |
| per-channel weights). Greedy WER (CER for zh) against the fp32 file, LibriSpeech |
| test-clean (150 clips) and FLEURS (80 per language): |
|
|
| | | en | de | es | pl | fr | zh | |
| | --- | --- | --- | --- | --- | --- | --- | |
| | fp32 | 6.99 | 35.82 | 22.93 | 63.73 | 54.10 | 29.57 | |
| | int8 | 7.05 | 35.60 | 21.61 | 63.32 | 54.20 | 28.64 | |
|
|
| ## Repository structure |
|
|
| ``` |
| config.json 34 B |
| coreml/config.json 1.5 kB |
| coreml/whisper_tiny_coreml_fp16.pte 72.9 MB |
| mlx/config.json 2.6 kB |
| mlx/whisper_tiny_mlx_bf16.pte 72.6 MB |
| mlx/whisper_tiny_mlx_int8.pte 57.1 MB |
| tokenizer.json 2.4 MB |
| vulkan/config.json 2.6 kB |
| vulkan/whisper_tiny_vulkan_fp16.pte 204 MB |
| vulkan/whisper_tiny_vulkan_int8.pte 124 MB |
| xnnpack/config.json 2.6 kB |
| xnnpack/whisper_tiny_xnnpack_fp32.pte 222 MB |
| xnnpack/whisper_tiny_xnnpack_int8.pte 175 MB |
| ``` |
|
|
| ## Compatibility |
|
|
| These files are published for the **ExecuTorch v1.4.1** runtime. ExecuTorch |
| gives no forward compatibility guarantee, so an older runtime may fail to load |
| them. |
|
|
| To use them in React Native ExecuTorch, pass the model constant shipped in the |
| library's model registry to the corresponding task pipeline. See the |
| [documentation](https://docs.swmansion.com/react-native-executorch/docs/fundamentals/downloading-models). |
|
|
| To load these files in your own ExecuTorch runtime, read the |
| [compatibility note](https://github.com/pytorch/executorch/blob/main/runtime/COMPATIBILITY.md) |
| first. |
|
|
