xbl916/whispercpp repository overview

services:
  sdforge:
    container_name: whisper
    image: whisper:cudap4
    runtime: nvidia
    shm_size: 128g
    ports:
      - 0.0.0.0:9996:8080
    environment:
      - PUID=0
      - GUID=0
      - OMP_NUM_THREADS=4
      - MKL_THREADING_LAYER=INTE
      - MKL_SERVICE_FORCE_INTEL=1
      - TZ=Asia/Shanghai
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
      - NVIDIA_VISIBLE_DEVICES=1
    volumes:
      - /root/model/:/data2
    ulimits:
      memlock:
        soft: -1
        hard: -1
    restart: unless-stopped
    #      --log-verbose 详细日志  --batch-size 300 批大小 -dev cuda0,cuda1  -sm row --ctx-size 122880
    command: >
      "./build/bin/whisper-server --model /data2/ggml-large-v3-turbo-q5_0.bin --host 0.0.0.0 --port 8080 --vad --vad-model /data2/ggml-silero-v5.1.2.bin -l auto --inference-path /v1/audio/transcriptions --convert" ```