d08bf5844a
- Rename deployment to rtx6000-brain-vllm to match scaler target ref - Set deployment replicas to 0 (KEDA controlled) - Add pollingInterval: 10s and cooldownPeriod: 30s for faster response This enables scale-to-zero when idle and quick scale-up on first request. See https://github.com/sirius0xdev/gcloud-lab/tree/master/infrastructure%2Fgpus%2Fbase
141 lines
3.5 KiB
YAML
141 lines
3.5 KiB
YAML
|
|
apiVersion: apps/v1
|
|
kind: Deployment
|
|
metadata:
|
|
name: rtx6000-brain-vllm
|
|
namespace: customer1
|
|
labels:
|
|
app: rtx6000-brain
|
|
spec:
|
|
replicas: 0
|
|
selector:
|
|
matchLabels:
|
|
app: rtx6000-brain
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: rtx6000-brain
|
|
spec:
|
|
nodeSelector:
|
|
cloud.google.com/gke-accelerator: "nvidia-rtx-pro-6000"
|
|
tolerations:
|
|
- key: "nvidia.com/gpu-nvidia-rtx-pro-6000"
|
|
operator: "Equal"
|
|
value: "present"
|
|
effect: "NoSchedule"
|
|
containers:
|
|
- name: vllm-brain
|
|
image: vllm/vllm-openai:latest
|
|
command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
|
|
env:
|
|
- name: HF_TOKEN
|
|
value: ""
|
|
- name: TOKENIZER_MODE
|
|
value: "hf"
|
|
args:
|
|
- --model=Youssofal/Qwen3.6-27B-Abliterated-Heretic-Uncensored-BF16
|
|
- --host=0.0.0.0
|
|
- --port=8000
|
|
- --tensor-parallel-size=1
|
|
- --tokenizer-mode=hf
|
|
- --gpu-memory-utilization=0.95
|
|
- --kv-cache-dtype=fp8
|
|
- --max-model-len=131072
|
|
- --enable-auto-tool-choice
|
|
- --enable-chunked-prefill
|
|
- --max-num-batched-tokens=8192
|
|
- --max-num-seqs=4
|
|
- --trust-remote-code
|
|
- --dtype=auto
|
|
- --enable-prefix-caching
|
|
- --tool-call-parser=qwen3_xml
|
|
- --reasoning-parser=qwen3
|
|
- --disable-custom-all-reduce
|
|
ports:
|
|
- containerPort: 8000
|
|
resources:
|
|
limits:
|
|
nvidia.com/gpu: 1
|
|
memory: "100Gi"
|
|
|
|
requests:
|
|
nvidia.com/gpu: 1
|
|
memory: "80Gi"
|
|
cpu: "8"
|
|
|
|
startupProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8000
|
|
initialDelaySeconds: 30
|
|
periodSeconds: 10
|
|
timeoutSeconds: 10
|
|
failureThreshold: 60
|
|
successThreshold: 1
|
|
|
|
livenessProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8000
|
|
initialDelaySeconds: 180
|
|
periodSeconds: 15
|
|
failureThreshold: 5
|
|
successThreshold: 1
|
|
timeoutSeconds: 10
|
|
|
|
|
|
readinessProbe:
|
|
httpGet:
|
|
path: /v1/models
|
|
port: 8000
|
|
initialDelaySeconds: 120
|
|
periodSeconds: 10
|
|
timeoutSeconds: 15
|
|
failureThreshold: 8
|
|
successThreshold: 1
|
|
|
|
volumeMounts:
|
|
- name: model-cache
|
|
mountPath: /root/.cache/huggingface
|
|
volumes:
|
|
- name: model-cache
|
|
persistentVolumeClaim:
|
|
claimName: vllm-model-qwen3.6-27b-uncensored
|
|
---
|
|
apiVersion: storage.k8s.io/v1
|
|
kind: StorageClass
|
|
metadata:
|
|
name: hyperdisk-balanced
|
|
provisioner: pd.csi.storage.gke.io
|
|
volumeBindingMode: WaitForFirstConsumer
|
|
allowVolumeExpansion: true
|
|
parameters:
|
|
type: hyperdisk-balanced
|
|
---
|
|
apiVersion: v1
|
|
kind: PersistentVolumeClaim
|
|
metadata:
|
|
name: vllm-model-qwen3.6-27b-uncensored
|
|
namespace: customer1
|
|
spec:
|
|
accessModes:
|
|
- ReadWriteOnce
|
|
storageClassName: hyperdisk-balanced
|
|
resources:
|
|
requests:
|
|
storage: 100Gi
|
|
|
|
---
|
|
apiVersion: v1
|
|
kind: Service
|
|
metadata:
|
|
name: rtx6000-brain-service
|
|
namespace: customer1
|
|
spec:
|
|
selector:
|
|
app: rtx6000-brain
|
|
ports:
|
|
- protocol: TCP
|
|
port: 8000
|
|
targetPort: 8000
|
|
type: ClusterIP
|