runtime-llamacpp-v0.2.5

Prebuilt self-contained binaries for the FunASR llama.cpp / GGUF runtime: SenseVoice, Paraformer and Fun-ASR-Nano with built-in FSMN-VAD. Download the default quantized model with `bash download-funasr-model.sh <sensevoice|paraformer|nano>` (the helper requires the Hugging Face CLI: `pip install -U huggingface_hub`), then run `llama-funasr-cli` / `llama-funasr-sensevoice` / `llama-funasr-paraformer`.

## v0.2.5 boundary

This release fixes Vulkan host-to-device weight upload so model tensors are copied into the backend buffer before graph execution. The exact release commit is `f371370d4c5e4c61d13d4eb9c55cda2f4dd95e4f`. Local Linux validation passed for Vulkan llvmpipe with Q8 and F16 weights.

This does **not** establish that the reported AMD Windows hardware crash is fixed. Windows AMD users should use `funasr-llamacpp-windows-x64-vulkan.zip`, capture the new initialization-boundary logs, and report the GPU, driver version, model/quantization, command, and complete output on [FunASR #3479](https://github.com/modelscope/FunASR/issues/3479).

## Platform guidance

- Use the default x64 asset for maximum CPU compatibility.
- Use x64-avx2 on CPUs with AVX2/FMA/F16C/BMI2 for higher throughput.
- Vulkan assets are `linux-x64-vulkan` and `windows-x64-vulkan`; they require a working Vulkan driver/ICD and enable SenseVoiceSmall graph execution with `llama-funasr-sensevoice ... --backend vulkan`.
- The Windows CUDA asset targets CUDA architecture 86 and requires a compatible NVIDIA driver. Build from source for other GPU architectures.
- Build from source with `-DGGML_VULKAN=ON` to validate platform-specific GPU stacks.

No Python ASR runtime or local build is required. [Runtime documentation](https://github.com/modelscope/FunASR/blob/runtime-llamacpp-v0.2.5/runtime/llama.cpp/README.md)

## SHA256

```text
4350ed49b1220374231ddafc84fed2dbe6a1d5e19d830edcb6a252d19d78b5db  funasr-llamacpp-linux-arm64.tar.gz
f9bb3fe78776ef86feb334b02488d3b1a5c16926a9a1cc79b3d5b1219ef1fb30  funasr-llamacpp-linux-x64-avx2.tar.gz
88cd52ba1ce92af9bf344944521a9d47ed9ceb424b312a5993369228c1e8c4fb  funasr-llamacpp-linux-x64-vulkan.tar.gz
69d47bf724216093bedc89a50efe0a97c80a2f4b89367378a72d2d3d94e4f241  funasr-llamacpp-linux-x64.tar.gz
c6b3cea2c5238ae48175c931365231b1c47351f95ca289455536de0a959648af  funasr-llamacpp-macos-arm64.tar.gz
6f19b8af0ce623122c767a0160bb8faabbc9666ae4f0cef1675cdf6250a9ea26  funasr-llamacpp-windows-x64-avx2.zip
3bb704e6b86f2e2e613aba84c361a51273e729c107183c7b69a1f2781c95cce8  funasr-llamacpp-windows-x64-cuda.zip
d51e34afed66c0898dba3d944cb17d2abc92c5fb7758fa4ee376561d5ce35a4c  funasr-llamacpp-windows-x64-vulkan.zip
dc90e85bc9a22477c5a8b556427fa5a6e0caf8951b5b6ec0237317406ce11bf5  funasr-llamacpp-windows-x64.zip
```