Starstruk Logo MarkStarstruk

v1.4.2

modelscope/FunASRv1.4.2Aug 14, 2026by github-actions[bot]

AI Summary

This release focuses on stabilizing training workflows by fixing DeepSpeed initialization and gradient accumulation synchronization issues. It introduces new SRT output support for the llama.cpp runtime and updates deployment documentation for SenseVoice and vLLM. Additionally, the release bundles the llama.cpp runtime v0.2.0 for multiple operating systems.

Key Highlights

  • Add SRT output support for llama.cpp runtime
  • Release llama.cpp runtime v0.2.0 with binaries for Linux, macOS, and Windows
  • Fix DDP no_sync synchronization with gradient accumulation
  • Fix DeepSpeed mode initialization from top-level config
  • Update documentation for SenseVoice audio.cpp and native server deployment

New Features

  • SRT output format support in llama.cpp runtime
  • llama.cpp runtime v0.2.0 binaries (AVX2, Vulkan, CUDA variants for Linux and Windows, ARM64 for Linux and macOS)
  • Native SenseVoice server deployment documentation
  • SenseVoice audio.cpp deployment documentation
  • vLLM 0.27.1 compatibility validation

Full Release Notes

## What's Changed
* Fix long subtitle segments after punctuation alignment by @LauraGPT in https://github.com/modelscope/FunASR/pull/3469
* fix(train): align DDP no_sync with gradient accumulation by @YuCeong-May in https://github.com/modelscope/FunASR/pull/3472
* fix(train): initialize DeepSpeed mode from top-level config by @YuCeong-May in https://github.com/modelscope/FunASR/pull/3471
* docs(site): refresh ecosystem evidence by @LauraGPT in https://github.com/modelscope/FunASR/pull/3473
* test(site): refresh ecosystem contract by @LauraGPT in https://github.com/modelscope/FunASR/pull/3474
* docs(vllm): clarify official model paths by @LauraGPT in https://github.com/modelscope/FunASR/pull/3476
* llama.cpp: add srt output by @gumblex in https://github.com/modelscope/FunASR/pull/3480
* fix(vulkan): refresh ggml for AMD submission handling by @LauraGPT in https://github.com/modelscope/FunASR/pull/3482
* docs(site): publish llama.cpp v0.2.0 downloads by @LauraGPT in https://github.com/modelscope/FunASR/pull/3483
* docs: surface llama.cpp runtime v0.2.0 by @LauraGPT in https://github.com/modelscope/FunASR/pull/3484
* chore: prepare FunASR 1.4.2 release by @LauraGPT in https://github.com/modelscope/FunASR/pull/3485
* docs(site): add SenseVoice audio.cpp deployment by @LauraGPT in https://github.com/modelscope/FunASR/pull/3487
* fix(site): prevent mobile evidence overflow by @LauraGPT in https://github.com/modelscope/FunASR/pull/3488
* docs(site): add SenseVoice native server deployment by @LauraGPT in https://github.com/modelscope/FunASR/pull/3489
* docs(site): mark audio.cpp SenseVoice as merged to main by @LauraGPT in https://github.com/modelscope/FunASR/pull/3490
* docs: validate native FunASR on vLLM 0.27.1 by @LauraGPT in https://github.com/modelscope/FunASR/pull/3492
* docs: clarify llama.cpp speaker diarization support by @LauraGPT in https://github.com/modelscope/FunASR/pull/3491

## New Contributors
* @gumblex made their first contribution in https://github.com/modelscope/FunASR/pull/3480

**Full Changelog**: https://github.com/modelscope/FunASR/compare/v1.4.1...v1.4.2

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

This Python release pairs with the current prebuilt llama.cpp / GGUF runtime release: [runtime-llamacpp-v0.2.0](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.0).

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/v1.4.2/funasr-llamacpp-linux-arm64.tar.gz) | `c78987b2384c6aef339aea1bcd0e130070455d6394fa7ab7ca26840ead10d5da` |
| Linux x64 AVX2 | [funasr-llamacpp-linux-x64-avx2.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-linux-x64-avx2.tar.gz) | `02e10e9a46ea76a040c45d431efe51a3324e64f08c24d38e18c8a4d2781490cd` |
| Linux x64 Vulkan | [funasr-llamacpp-linux-x64-vulkan.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-linux-x64-vulkan.tar.gz) | `caf71b8c0b4c3249cebc4175e5406d3c588eb9e8966a00d571d4cc5070405385` |
| Linux x64 portable | [funasr-llamacpp-linux-x64.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-linux-x64.tar.gz) | `15e6407143b4fb91d90bb37f2a41c64c4d48ea0fbe6404b88a9b70269c84f240` |
| macOS arm64 | [funasr-llamacpp-macos-arm64.tar.gz](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-macos-arm64.tar.gz) | `416cbb289e31cb7575365d382155074e922fd061807a37b9ca0247dabd9bc6f9` |
| Windows x64 AVX2 | [funasr-llamacpp-windows-x64-avx2.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-windows-x64-avx2.zip) | `4db0f11f603c324a63545cd7009cdd45bb45576efe282cec22796b5fd42d8ea1` |
| Windows x64 CUDA | [funasr-llamacpp-windows-x64-cuda.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-windows-x64-cuda.zip) | `7f2f9ef4d7e0291b284a295ec74bbeca9ea635a7f5f42d0ad06eb780c0d6efc1` |
| Windows x64 Vulkan | [funasr-llamacpp-windows-x64-vulkan.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-windows-x64-vulkan.zip) | `90b45240c6ccc9177c25490a11848de60a406e129391c8736b14521c0c28cdcb` |
| Windows x64 portable | [funasr-llamacpp-windows-x64.zip](https://github.com/modelscope/FunASR/releases/download/v1.4.2/funasr-llamacpp-windows-x64.zip) | `297c962346d7e30d7a7c2c860dfaab3ff07d01fddf15e6fc5212ca9545441a51` |

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==1.4.2"
```
<!-- funasr-runtime-downloads:end -->