Select a model and protocol before selecting a container or binary. Start at the deployment matrix for versioned commands, tested hardware and known limits. Older release notes are preserved in release history; they are not a current capacity promise.

Choose a serving path

Need	Entry point	Boundary
Python HTTP transcription	OpenAI-compatible server	Application API compatibility is separate from model accuracy or realtime support.
Fun-ASR-Nano decoder acceleration	vLLM guide	Native vLLM and FunASR split-engine have different checkpoint/layout and API contracts.
Local portable GGUF inference	llama.cpp	Use the platform/backend package and matching GGUF model; build success is not every-device validation.
Native ONNX CPU inference	ONNX Runtime	See the JSONL/timestamp output contract.
Offline unified transcription and diarization	MOSS-Transcribe-Diarize	Third-party OpenMOSS model; offline anonymous speaker labels, not realtime or known-person identity.
Long-lived streaming or two-pass sessions	C++ WebSocket protocol	Do not send OpenAI HTTP requests or another runtime's WebSocket messages to this endpoint.
Cluster-managed private HTTP service	Kubernetes templates	Configure resources, persistent cache, probes, upload limits and gateway policy for the target cluster.

File Transcription Service, Mandarin (GPU)

Follow the GPU development guide. The guide covers the native runtime stack; it is not an installation recipe for every model in the Model Zoo. Benchmark your exact image, weights and GPU.

File Transcription Service, English (CPU)

Use the English tutorial and advanced configuration. Select the English checkpoint rather than inferring language coverage from the container name alone.

Real-time Transcription Service, Mandarin (CPU)

Use the streaming tutorial, then verify the protocol and matching multi-client example. Validate sample rate, chunking, finalization, reconnect behavior and session isolation with actual audio. C++ two-pass and Fun-ASR-Nano Python streaming are different implementations. The separate Nano realtime benchmark uses Nano's START/STOP protocol and must not be used against the C++ endpoint.

File Transcription Service, Mandarin (CPU)

Use the offline tutorial and advanced configuration. Model selection explains when Paraformer, SenseVoice or a different recognizer fits the workload.

Client and platform adapters

Python WebSocket, Python HTTP, Java, Go.
Browser client, gRPC, Triton.
Android and iOS are separate porting guides, not a guarantee that each desktop release supports those devices.

Follow each adapter's protocol and dependency instructions. An SDK example is not automatically a supported production package for every target platform.

Production checklist

Record the exact commit/image digest, model revision, configuration and target hardware.
Run a known-audio transcription and inspect the raw output, not only a health endpoint.
Measure representative quality, latency, concurrency, memory and failure behavior.
Add authentication, TLS, request limits and privacy controls using the security guide.
Preserve the previous model/artifact/configuration and exercise rollback.
Report unresolved failures with the troubleshooting checklist; published code is not proof that a reporter's hardware issue is resolved.

Historical releases

The full release history preserves earlier Docker tags, dates and benchmark references. For a new deployment, use a current deployment manual and its explicit test boundary.