diff --git a/whisper/compose.yaml b/whisper/compose.yaml index b9d6807..d53ffad 100644 --- a/whisper/compose.yaml +++ b/whisper/compose.yaml @@ -1,25 +1,23 @@ services: whisper-live: - image: hwdsl2/whisper-live-server:latest + image: ghcr.io/collabora/whisperlive-openvino:latest container_name: whisper-live restart: unless-stopped ports: - - "9097:9090" # WebSocket — real-time streaming clients - - "8071:8000" # REST API — OpenAI-compatible /v1/audio/transcriptions + - "9097:9090" # Maps your existing client port 9097 to the official 9090 port volumes: - - whisper-live-data:/var/lib/whisper-live # persists downloaded models - env_file: - - whisper-live.env - # --- ADDED FOR INTEL iGPU PASSTHROUGH --- + - whisper-live-data:/var/lib/whisper-live + + # Tells WhisperLive to use the OpenVINO hardware engine + command: ["python3", "run_server.py", "--port", "9090", "--backend", "openvino"] + devices: - /dev/dri/renderD128:/dev/dri/renderD128 - - /dev/dri/card1:/dev/dri/card1 - # ---------------------------------------- + - /dev/dri/card1:/dev/dri/card1 # Change to card0 if your 13700K is the only GPU - # The 13700K has 8 P-cores; give the container access to all of them. - # Adjust cpuset if you want to reserve cores for other workloads. - cpuset: "0-15" # all 16 logical P-core threads - mem_limit: 6g # medium int8 needs ~2 GB; headroom for concurrency + cpuset: "0-15" # Giving it all 16 P-core logical threads + mem_limit: 6g volumes: whisper-live-data: +