FunASR deployment hub: vLLM, llama.cpp, OpenAI API, realtime, and CPU

[LauraGPT]

FunASR deployment hub is live

The new bilingual FunASR deployment hub helps teams choose and operate a runtime instead of treating every deployment as the same workload.

Path	Best fit	Guide
vLLM	Fun-ASR-Nano GPU batching and OpenAI-compatible transcription	Open guide
llama.cpp / GGUF	Standalone Linux, macOS, Windows, CPU, CUDA, and Vulkan packages	Open guide
OpenAI-compatible API	Private /v1/audio/transcriptions service for existing clients	Open guide
Realtime WebSocket	Online and two-pass captions with explicit final-result semantics	Open guide
Containers / Kubernetes	Reproducible private services with probes and persistent model cache	Open guide
ONNX / C++ CPU	Paraformer and SenseVoice services without a GPU requirement	Open guide

The home page includes a deployment selector based on workload, hardware, and operating priorities. Each route records the supported model/runtime combination, commands, a smoke test, operational and security boundaries, known limitations, and source evidence.

Benchmark policy

Public numbers are shown only with their available model, runtime, hardware, workload, audio, settings, timing scope, result, qualification, source, and verification date. Missing hardware or timing details are stated explicitly; those records are references, not capacity promises. Reproduce results on your target hardware and audio distribution before making a production decision.

Release verification

• 92 HTML pages passed static validation.
• 41 Python release and contract tests passed.
• 7 Playwright checks passed across mobile, tablet, desktop, and wide desktop viewports.
• Public smoke tests covered 25 pages, five fixed conversion routes, security/cache headers, and immutable hashed assets.
• TLS 1.2 and 1.3 were verified.
• A real rollback to the previous site release was exercised before restoring the product release.

中文入口：FunASR 工业部署中心；英文入口：English deployment hub。

Repository entry points now expose the hub from both README.md and README_zh.md.