Compatibility

Compatibility

Hardware and runtime

Target	Status	Notes
NVIDIA RTX 3000 → 5000 (x86_64, CUDA cu128)	Supported	Single cu128 image, compute capability sm_86–sm_120.
NVIDIA Jetson Orin (arm64)	Supported	JetPack 6 / L4T r36 (CUDA 12.6, cuDNN 9.3); use the Jetson image.
CPU (x86_64)	Supported	`USE_GPU=false`; reduced throughput.
Apple Silicon (MPS)	Experimental	Auto-detected for local runs; not a published container target.
AMD (ROCm), Intel	Planned	Not currently supported.

See Hardware profiles for image selection.

Container engines and orchestration

Docker / OCI engines with the NVIDIA Container Toolkit for GPU access.
Docker Compose (files provided) for single-host operation.
Kubernetes 1.25+ with the NVIDIA device plugin / GPU Operator; Helm 3.8+ (OCI) for the chart.

API compatibility

OpenAI-compatible /v1/audio/speech, /v1/audio/voices, and /v1/models — usable with the OpenAI SDKs by setting base_url to .../v1.
Standard OpenAI voice names are accepted and mapped to designed voices. See Voices and languages.
Extended, non-OpenAI endpoints are under /dev and /debug. See Extended API.

Audio formats

Format	Supported	Notes
`mp3`	Yes	Default.
`wav`	Yes	16-bit PCM RIFF.
`opus`	Yes	Ogg/Opus.
`flac`	Yes	Lossless.
`pcm`	Yes	Raw signed 16-bit LE, 24 kHz, mono.
`aac`	No	Accepted by the request schema but not currently produced.

All output is 24 kHz mono. See Streaming and audio formats.

Runtime versions

Python 3.10 in the images.
PyTorch cu128 on x86_64; a JetPack-6-matched PyTorch build on Jetson.

The authoritative dependency set is pyproject.toml.