gpu.jsonc

{
  "$schema": "https://gpu-cli.sh/schema/v1/gpu.json",
  "project_id": "gpu-tts",

  "gpu_types": [
    { "type": "RTX 4090" },
    { "type": "RTX A6000" }
  ],
  "min_vram": 24,

  // Start combined TTS + STT server (deps installed by GPU CLI from requirements.txt)
  // PyAV handles WebM/Opus decoding — no system ffmpeg needed
  "startup": "find /gpu-cli-workspaces -name '__pycache__' -type d -exec rm -rf {} + 2>/dev/null; fuser -k -9 8000/tcp 2>/dev/null; sleep 2; fuser -k -9 8000/tcp 2>/dev/null; sleep 1; TTS_MODEL=${TTS_MODEL:-chatterbox} STT_MODEL=${STT_MODEL:-large-v3-turbo} python -B server.py",

  "ports": [
    { "port": 8000, "http": { "activity_paths": ["/tts/tts", "/stt/stt"] } }
  ],
  "health_check_paths": ["/health", "/tts/health", "/stt/health"],
  "persistent_proxy": false,
  "keep_alive_minutes": 10,

  "outputs": ["output/"],

  "template": {
    "name": "GPU TTS + STT",
    "description": "Voice cloning TTS (Chatterbox) and transcription STT (faster-whisper) on cloud GPUs",
    "author": "jameslal"
  },
  "inputs": [
    {
      "type": "select",
      "key": "TTS_MODEL",
      "label": "TTS Model",
      "description": "Chatterbox standard or turbo (faster, lower latency)",
      "options": [
        { "label": "Standard", "value": "chatterbox" },
        { "label": "Turbo", "value": "chatterbox-turbo" }
      ]
    },
    {
      "type": "select",
      "key": "STT_MODEL",
      "label": "STT Model",
      "description": "Whisper model size (larger = more accurate, slower)",
      "options": [
        { "label": "Large V3 Turbo", "value": "large-v3-turbo" },
        { "label": "Large V3", "value": "large-v3" },
        { "label": "Medium", "value": "medium" },
        { "label": "Small", "value": "small" },
        { "label": "Base", "value": "base" },
        { "label": "Tiny", "value": "tiny" }
      ]
    }
  ],
  "volume_mode": "global"
}