Initial
This commit is contained in:
parent
fb3152e70b
commit
bd4ccbe7df
|
|
@ -0,0 +1,85 @@
|
||||||
|
# docker-compose.yml – Hermes Agent + vLLM (Qwen3-14B-AWQ)
|
||||||
|
# Host: Cortex (eigenständiger Server, 2x RTX 3060 12GB)
|
||||||
|
#
|
||||||
|
# Beide Services laufen im selben Docker-Netz "cortex-net" auf diesem Host.
|
||||||
|
# Hermes spricht vLLM intern über http://vllm:8000/v1 an – kein extra
|
||||||
|
# Netzwerk-Routing nötig, alles lokal auf Cortex.
|
||||||
|
#
|
||||||
|
|
||||||
|
services:
|
||||||
|
vllm:
|
||||||
|
image: vllm/vllm-openai:latest
|
||||||
|
container_name: cortex-vllm
|
||||||
|
restart: unless-stopped
|
||||||
|
runtime: nvidia
|
||||||
|
ipc: host
|
||||||
|
ports:
|
||||||
|
- "8005:8000"
|
||||||
|
environment:
|
||||||
|
- NVIDIA_VISIBLE_DEVICES=0,1
|
||||||
|
- HUGGING_FACE_HUB_TOKEN=${HF_TOKEN:-}
|
||||||
|
volumes:
|
||||||
|
- ${HOME:-/root}/.cache/huggingface:/root/.cache/huggingface
|
||||||
|
deploy:
|
||||||
|
resources:
|
||||||
|
reservations:
|
||||||
|
devices:
|
||||||
|
- driver: nvidia
|
||||||
|
count: 2
|
||||||
|
capabilities: [gpu]
|
||||||
|
command:
|
||||||
|
- --model=Qwen/Qwen3-14B-AWQ
|
||||||
|
- --quantization=awq
|
||||||
|
- --tensor-parallel-size=2
|
||||||
|
- --gpu-memory-utilization=0.90
|
||||||
|
- --max-model-len=32768
|
||||||
|
- --hf-overrides={"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":32768}}
|
||||||
|
- --served-model-name=qwen3-14b-awq
|
||||||
|
networks:
|
||||||
|
- cortex-net
|
||||||
|
|
||||||
|
hermes-gateway:
|
||||||
|
image: nousresearch/hermes-agent:latest
|
||||||
|
container_name: cortex-hermes-gateway
|
||||||
|
restart: unless-stopped
|
||||||
|
depends_on:
|
||||||
|
- vllm
|
||||||
|
command: ["gateway", "run"]
|
||||||
|
shm_size: 1g
|
||||||
|
ports:
|
||||||
|
- "8642:8642"
|
||||||
|
volumes:
|
||||||
|
- ${HOME:-/root}/.hermes:/opt/data
|
||||||
|
environment:
|
||||||
|
- HERMES_UID=${HERMES_UID:-10000}
|
||||||
|
- HERMES_GID=${HERMES_GID:-10000}
|
||||||
|
# Zeigt Hermes auf den lokalen vLLM-Server
|
||||||
|
- OPENAI_BASE_URL=http://vllm:8000/v1
|
||||||
|
- OPENAI_API_KEY=dummy
|
||||||
|
# Gateway-API nur aktivieren, wenn du sie extern brauchst:
|
||||||
|
# - API_SERVER_HOST=0.0.0.0
|
||||||
|
# - API_SERVER_KEY=${API_SERVER_KEY}
|
||||||
|
networks:
|
||||||
|
- cortex-net
|
||||||
|
|
||||||
|
hermes-dashboard:
|
||||||
|
image: nousresearch/hermes-agent:latest
|
||||||
|
container_name: cortex-hermes-dashboard
|
||||||
|
restart: unless-stopped
|
||||||
|
depends_on:
|
||||||
|
- hermes-gateway
|
||||||
|
command: ["dashboard", "--host", "0.0.0.0", "--no-open"]
|
||||||
|
ports:
|
||||||
|
- "127.0.0.1:9119:9119" # nur lokal auf Cortex, kein LAN-Expose
|
||||||
|
volumes:
|
||||||
|
- ${HOME:-/root}/.hermes:/opt/data
|
||||||
|
environment:
|
||||||
|
- HERMES_UID=${HERMES_UID:-10000}
|
||||||
|
- HERMES_GID=${HERMES_GID:-10000}
|
||||||
|
- GATEWAY_HEALTH_URL=http://cortex-hermes-gateway:8642
|
||||||
|
networks:
|
||||||
|
- cortex-net
|
||||||
|
|
||||||
|
networks:
|
||||||
|
cortex-net:
|
||||||
|
driver: bridge
|
||||||
Loading…
Reference in New Issue