33 lines
876 B
YAML
33 lines
876 B
YAML
services:
|
|
bert-all-services:
|
|
build:
|
|
context: .
|
|
dockerfile: Dockerfile
|
|
container_name: bert-all-services
|
|
user: "0:0" # 显式使用 root
|
|
ports:
|
|
- "5002-5010:5002-5010"
|
|
env_file:
|
|
- .env
|
|
environment:
|
|
- DEVICE=${DEVICE:-cuda}
|
|
- MAX_WORKERS=${MAX_WORKERS:-4}
|
|
- BATCH_SIZE=${BATCH_SIZE:-32}
|
|
- NVIDIA_VISIBLE_DEVICES=${NVIDIA_VISIBLE_DEVICES:-all}
|
|
- NVIDIA_DRIVER_CAPABILITIES=${NVIDIA_DRIVER_CAPABILITIES:-all}
|
|
- MODEL_CACHE_DIR=/app/model_cache
|
|
volumes:
|
|
- ./logs:/app/logs
|
|
- ./models:/app/models
|
|
- ./model_cache:/app/model_cache
|
|
restart: unless-stopped
|
|
deploy:
|
|
resources:
|
|
limits:
|
|
memory: 8G
|
|
reservations:
|
|
memory: 4G
|
|
devices:
|
|
- driver: nvidia
|
|
count: all
|
|
capabilities: [gpu] |