|
@@ -12,6 +12,13 @@ services:
|
|
|
--api-port 8000
|
|
|
volumes:
|
|
|
- ./models:/root/.cache/huggingface
|
|
|
+ deploy:
|
|
|
+ resources:
|
|
|
+ reservations:
|
|
|
+ devices:
|
|
|
+ - driver: nvidia
|
|
|
+ count: all
|
|
|
+ capabilities: [ gpu ]
|
|
|
|
|
|
# # vLLM - Qwen3 4B
|
|
|
# vllm-4b:
|
|
@@ -41,7 +48,8 @@ services:
|
|
|
|
|
|
# Milvus 向量数据库
|
|
|
milvus:
|
|
|
- image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/milvusdb/milvus:v2.4.4
|
|
|
+# image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/milvusdb/milvus:v2.4.4
|
|
|
+ image: milvusdb/milvus:v2.4.4-standalone
|
|
|
container_name: milvus
|
|
|
ports:
|
|
|
- "19530:19530"
|
|
@@ -51,11 +59,4 @@ services:
|
|
|
- MINIO_USE_EMBED=true
|
|
|
- PULSAR_USE_EMBED=true
|
|
|
volumes:
|
|
|
- - ./milvus_data:/var/lib/milvus
|
|
|
- deploy:
|
|
|
- resources:
|
|
|
- reservations:
|
|
|
- devices:
|
|
|
- - driver: nvidia
|
|
|
- count: all
|
|
|
- capabilities: [ gpu ]
|
|
|
+ - ./milvus_data:/var/lib/milvus
|