composte-hermes.yml aktualisiert
This commit is contained in:
parent
80e23cfaf4
commit
3514e0a3e3
|
|
@ -1,7 +1,3 @@
|
||||||
# docker-compose.yml – Hermes Agent + vLLM (Qwen3-14B-AWQ)
|
|
||||||
# Host: Cortex (2x RTX 3060 12GB)
|
|
||||||
# Daten liegen unter /opt/vllm und /opt/hermes (Sicherung/Backup)
|
|
||||||
|
|
||||||
services:
|
services:
|
||||||
vllm:
|
vllm:
|
||||||
image: vllm/vllm-openai:latest
|
image: vllm/vllm-openai:latest
|
||||||
|
|
@ -10,7 +6,7 @@ services:
|
||||||
runtime: nvidia
|
runtime: nvidia
|
||||||
ipc: host
|
ipc: host
|
||||||
ports:
|
ports:
|
||||||
- "8005:8000"
|
- "8005:8005"
|
||||||
environment:
|
environment:
|
||||||
- NVIDIA_VISIBLE_DEVICES=0,1
|
- NVIDIA_VISIBLE_DEVICES=0,1
|
||||||
- HUGGING_FACE_HUB_TOKEN=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
|
- HUGGING_FACE_HUB_TOKEN=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
|
||||||
|
|
@ -26,6 +22,7 @@ services:
|
||||||
capabilities: [gpu]
|
capabilities: [gpu]
|
||||||
command:
|
command:
|
||||||
- --model=Qwen/Qwen3-14B-AWQ
|
- --model=Qwen/Qwen3-14B-AWQ
|
||||||
|
- --port=8005
|
||||||
- --quantization=awq
|
- --quantization=awq
|
||||||
- --tensor-parallel-size=2
|
- --tensor-parallel-size=2
|
||||||
- --gpu-memory-utilization=0.82
|
- --gpu-memory-utilization=0.82
|
||||||
|
|
@ -54,14 +51,14 @@ services:
|
||||||
condition: service_healthy
|
condition: service_healthy
|
||||||
shm_size: 1g
|
shm_size: 1g
|
||||||
ports:
|
ports:
|
||||||
- "9119:9119" # Dashboard, jetzt im LAN erreichbar
|
- "9119:9119"
|
||||||
- "8642:8642" # Gateway-API, im LAN erreichbar
|
- "8642:8642"
|
||||||
volumes:
|
volumes:
|
||||||
- /opt/hermes:/opt/data
|
- /opt/hermes:/opt/data
|
||||||
environment:
|
environment:
|
||||||
- HERMES_UID=${HERMES_UID:-10000}
|
- HERMES_UID=${HERMES_UID:-10000}
|
||||||
- HERMES_GID=${HERMES_GID:-10000}
|
- HERMES_GID=${HERMES_GID:-10000}
|
||||||
- OPENAI_BASE_URL=http://vllm:8000/v1
|
- OPENAI_BASE_URL=http://vllm:8005/v1
|
||||||
- OPENAI_API_KEY=dummy
|
- OPENAI_API_KEY=dummy
|
||||||
- GATEWAY_HEALTH_URL=http://localhost:8642
|
- GATEWAY_HEALTH_URL=http://localhost:8642
|
||||||
entrypoint: ["/bin/sh", "-c"]
|
entrypoint: ["/bin/sh", "-c"]
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue