Stable Audio 3 scales from a laptop to a GPU server.

Optimized inference runtimes are available under optimized/ — pick by platform:

| Route | Platforms | Backend | One-liner | |---|---|---|---| | optimized/tflite | macOS / Linux / Windows, x86 & ARM | CPU (LiteRT/XNNPACK) | curl -LsSf https://raw.githubusercontent.com/Stability-AI/stable-audio-3/main/optimized/tflite/bootstrap.sh \| bash<br>Windows: irm https://raw.githubusercontent.com/Stability-AI/stable-audio-3/main/optimized/tflite/bootstrap.ps1 \| iex <br>(the curl|bash line needs Git Bash or WSL on Windows — stock Windows has no bash) | | optimized/mlx | Apple Silicon Macs | Metal GPU (MLX) | curl -LsSf https://raw.githubusercontent.com/Stability-AI/stable-audio-3/main/optimized/mlx/bootstrap.sh \| bash | | optimized/tensorRT | Linux + NVIDIA GPU | CUDA/TensorRT | curl -LsSf https://raw.githubusercontent.com/Stability-AI/stable-audio-3/main/optimized/tensorRT/bootstrap.sh \| bash |

Beyond inference, optimized/mlx and optimized/tflite each ship a web UI (./sa3-gradio), and optimized/mlx also does LoRA training on Apple Silicon (pure-MLX, no PyTorch) — see its LoRA training section, or underfit for a full training dashboard built on it.