kubernetes-guide/codeblock/llama/llama3-gpu-70b.yaml

83 lines
2.1 KiB
YAML
Raw Normal View History

2024-04-26 15:40:36 +08:00
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: ollama
namespace: llama
spec:
serviceName: "ollama"
replicas: 1
selector:
matchLabels:
app: ollama
template:
metadata:
labels:
app: ollama
spec:
initContainers:
- name: pull
image: ollama/ollama:latest
tty: true
stdin: true
command:
- bash
- -c
- |
# highlight-next-line
model="llama3:70b" # 替换需要使用的模型,模型库列表: https://ollama.com/library/llama3
ollama serve &
sleep 5 # 等待 ollama server 就绪,就绪后才能执行 ollama cli 工具的命令
result=`ollama list | grep $model`
if [ "$result" == "" ]; then
echo "downloading model $model"
ollama pull $model
else
echo "model $model already been downloaded"
fi
volumeMounts:
- name: ollama-volume
mountPath: /root/.ollama
containers:
- name: ollama
image: ollama/ollama:latest
ports:
- containerPort: 11434
resources:
requests:
cpu: "2000m"
memory: "2Gi"
# highlight-next-line
nvidia.com/gpu: "1" # 声明使用一张 N 卡
limits:
cpu: "4000m"
memory: "4Gi"
volumeMounts:
- name: ollama-volume
mountPath: /root/.ollama
tty: true
volumeClaimTemplates:
- metadata:
name: ollama-volume
spec:
accessModes: ["ReadWriteOnce"]
resources:
requests:
# highlight-next-line
storage: 200Gi # 注意要确保磁盘容量能够容纳得下模型的体积
---
apiVersion: v1
kind: Service
metadata:
name: ollama
namespace: llama
labels:
app: ollama
spec:
type: ClusterIP
ports:
- port: 11434
protocol: TCP
targetPort: 11434
selector:
app: ollama