v1.4.5

modelscope/FunASRv1.4.5Aug 27, 2026by github-actions[bot]

AI Summary

This release simplifies the default Python inference installation by removing `torchaudio` as a hard dependency and introducing a `kaldi-native-fbank` backend. It also bundles prebuilt llama.cpp runtimes alongside the Python distributions to streamline deployment.

Key Highlights

Removed `torchaudio` as a hard dependency for standard inference.
Added optional `kaldi-native-fbank` backend for a smaller dependency path.
Bundled nine prebuilt llama.cpp runtime archives for Linux, macOS, and Windows.
Validated Ascend 910B performance achieving an RTF of 0.016.
Published wheels and source distributions on PyPI.

New Features

New optional `kaldi-native-fbank` backend via the `knf` extra.
Integrated GPU runtime backends support.
Prebuilt llama.cpp runtime v0.2.1 binaries for various platforms.
PyPI publication of source distributions and wheels.

Full Release Notes

## FunASR 1.4.5

This release makes the default Python inference install lighter and puts the current prebuilt llama.cpp / GGUF runtimes beside the Python distributions so deployment users can start from one release page.

### Install

```bash
python -m pip install -U "funasr==1.4.5"
```

Install the optional `kaldi-native-fbank` backend with:

```bash
python -m pip install -U "funasr[knf]==1.4.5"
```

### Highlights

- `torchaudio` is no longer a hard dependency for standard inference. Existing torchaudio-based workflows remain supported when torchaudio is installed explicitly.
- The optional `knf` extra adds `kaldi-native-fbank` for a smaller inference dependency path.
- Ascend 910B validation processed 70.47 seconds of audio in 1.15 seconds (RTF 0.016) under the documented test contract.
- Nine prebuilt llama.cpp / GGUF runtime archives from the pinned `runtime-llamacpp-v0.2.1` release are attached below for Linux, macOS, and Windows CPU/GPU deployment.
- The Python wheel, source distribution, and `SHA256SUMS-v1.4.5` are attached to this release and also published on [PyPI](https://pypi.org/project/funasr/1.4.5/).

### What's Changed

- Remove the hard torchaudio inference dependency and add the `kaldi-native-fbank` backend by @li-lizhe in https://github.com/modelscope/FunASR/pull/3526
- Allow integrated GPU runtime backends by @LauraGPT in https://github.com/modelscope/FunASR/pull/3531
- Publish llama.cpp runtime v0.2.1 downloads by @LauraGPT in https://github.com/modelscope/FunASR/pull/3532
- Prepare the 1.4.5 package and release contract by @LauraGPT in https://github.com/modelscope/FunASR/pull/3534

Thanks to @li-lizhe for their first contribution.

**Full Changelog**: https://github.com/modelscope/FunASR/compare/v1.4.4...v1.4.5

<!-- funasr-runtime-downloads:start -->
## Runtime downloads

This Python release pairs with the current prebuilt llama.cpp / GGUF runtime release: [runtime-llamacpp-v0.2.1](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.1).

The same verified runtime assets are attached directly to this Python release so users can find the package and self-contained `llama-funasr-*` binaries in one place.

| Platform | Asset | SHA-256 |
|---|---|---|
| Linux arm64 | [funasr-llamacpp-linux-arm64.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-linux-arm64.tar.gz) | `9657e519986a5db72082aab508439559987fd4e563733d11da4662f898c75c45` |
| Linux x64 AVX2 | [funasr-llamacpp-linux-x64-avx2.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-linux-x64-avx2.tar.gz) | `32e459673297fc01fcda6901ca7d63cc9e64896a43a3c9edcb7a8e1ef70fcd04` |
| Linux x64 Vulkan | [funasr-llamacpp-linux-x64-vulkan.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-linux-x64-vulkan.tar.gz) | `d45d3acb77e05c4bae94d818ed4a5c66852be87b22c52723bd576377c055c6e6` |
| Linux x64 portable | [funasr-llamacpp-linux-x64.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-linux-x64.tar.gz) | `1bc83dd36d11b1a2b9a45faaf77a5d1014000ac641c93dfec73d24d71ec51caf` |
| macOS arm64 | [funasr-llamacpp-macos-arm64.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-macos-arm64.tar.gz) | `bc63c4d4b96f2465f1d258600668a971f4f600d661f1859b03797cefaa417167` |
| Windows x64 AVX2 | [funasr-llamacpp-windows-x64-avx2.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-windows-x64-avx2.zip) | `67e428a91614f8a0c1d53dc2d79cce8efc7916c030ae2ddce2cbcf667ec1c502` |
| Windows x64 CUDA | [funasr-llamacpp-windows-x64-cuda.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-windows-x64-cuda.zip) | `af32e0ef352880fcfd5589e0152c44038bc81f462f42c7a6ca91a0d9879bac58` |
| Windows x64 Vulkan | [funasr-llamacpp-windows-x64-vulkan.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-windows-x64-vulkan.zip) | `5fd179c6aefc18477eb79fa6235c70d131a96eba621318bb84e901ae85035578` |
| Windows x64 portable | [funasr-llamacpp-windows-x64.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.5/funasr-llamacpp-windows-x64.zip) | `b8f2b8f241b57921d82d64068d9b5695629779f3db5f3205a730cb3810232bb4` |

Quick start: download one asset, unpack it, then run the bundled `download-funasr-model.sh <sensevoice|paraformer|nano>` helper and one of `llama-funasr-cli`, `llama-funasr-sensevoice`, or `llama-funasr-paraformer`.

For Python users, install from PyPI:

```bash
python -m pip install -U "funasr==1.4.5"
```
<!-- funasr-runtime-downloads:end -->