- Chunk Text: word-level fallback for sentences > 600 chars - watch.py: keep MP3 after transcription - Respond Accepted: remove cross-node reference - rocinante/: Whisper CUDA + watcher stack for RTX 4090 transcription
41 lines
854 B
YAML
41 lines
854 B
YAML
name: rocinante
|
|
|
|
services:
|
|
|
|
whisper:
|
|
image: fedirz/faster-whisper-server:latest-cuda
|
|
container_name: rocinante-whisper
|
|
restart: unless-stopped
|
|
ports:
|
|
- "8787:8000"
|
|
environment:
|
|
WHISPER__MODEL: large-v3
|
|
WHISPER__INFERENCE_DEVICE: cuda
|
|
TZ: ${TZ}
|
|
deploy:
|
|
resources:
|
|
reservations:
|
|
devices:
|
|
- driver: nvidia
|
|
count: 1
|
|
capabilities: [gpu]
|
|
volumes:
|
|
- whisper-cache:/root/.cache/huggingface
|
|
|
|
watcher:
|
|
build: ./watcher
|
|
container_name: rocinante-watcher
|
|
restart: unless-stopped
|
|
environment:
|
|
WHISPER_URL: http://whisper:8000
|
|
N8N_WEBHOOK_URL: ${N8N_WEBHOOK_URL}
|
|
WATCH_DIR: /recordings
|
|
TZ: ${TZ}
|
|
volumes:
|
|
- C:/Recordings:/recordings
|
|
depends_on:
|
|
- whisper
|
|
|
|
volumes:
|
|
whisper-cache:
|