Story Crater Bot
50d00ae350
Frees verifier's GPU from an underused vLLM PRM deployment. qwen2.5:3b-instruct moves off ornith-predictor's shared pod onto its own dedicated GPU (grm.yaml), so verification/judge traffic stops contending with ornith:35b's agent traffic. /v1/qwen/chat/completions now points at grm-predictor; path unchanged.
97 lines
2.5 KiB
YAML
97 lines
2.5 KiB
YAML
apiVersion: serving.kserve.io/v1beta1
|
|
kind: InferenceService
|
|
metadata:
|
|
annotations:
|
|
serving.kserve.io/deploymentMode: RawDeployment
|
|
# Same Kong-timeout rationale as ornith.yaml: these configure the Service
|
|
# KServe generates, not the Ingress, and matter once OLLAMA_KEEP_ALIVE=-1
|
|
# stops covering a cold load after a pod restart.
|
|
konghq.com/connect-timeout: "10000"
|
|
konghq.com/read-timeout: "3600000"
|
|
konghq.com/write-timeout: "3600000"
|
|
labels:
|
|
app.kubernetes.io/name: llm-grm
|
|
app.kubernetes.io/part-of: llm-serving
|
|
name: grm
|
|
namespace: llm-serving
|
|
spec:
|
|
predictor:
|
|
containers:
|
|
- command:
|
|
- /bin/sh
|
|
- -c
|
|
- 'set -e
|
|
|
|
ollama serve &
|
|
|
|
SERVE_PID=$!
|
|
|
|
until ollama list >/dev/null 2>&1; do sleep 2; done
|
|
|
|
ollama pull qwen2.5:3b-instruct
|
|
|
|
ollama run qwen2.5:3b-instruct "ok" >/dev/null 2>&1 || true
|
|
|
|
wait $SERVE_PID
|
|
|
|
'
|
|
env:
|
|
- name: OLLAMA_HOST
|
|
value: 0.0.0.0:8080
|
|
- name: OLLAMA_MODELS
|
|
value: /mnt/models/ollama
|
|
- name: OLLAMA_CONTEXT_LENGTH
|
|
value: '32768'
|
|
- name: OLLAMA_KEEP_ALIVE
|
|
value: '-1'
|
|
- name: OLLAMA_NUM_PARALLEL
|
|
value: '1'
|
|
# Room for a second verification/reward model alongside qwen2.5:3b
|
|
# without a redeploy -- matches ornith.yaml's pattern, one dedicated
|
|
# GPU now free for it instead of contending with ornith:35b's.
|
|
- name: OLLAMA_MAX_LOADED_MODELS
|
|
value: '2'
|
|
image: ollama/ollama:0.32.9@sha256:1685741456770df6e3cceb2a945a5f75e020f658d1701509668d6f4688f1dd3f
|
|
name: kserve-container
|
|
ports:
|
|
- containerPort: 8080
|
|
protocol: TCP
|
|
readinessProbe:
|
|
exec:
|
|
command:
|
|
- /bin/sh
|
|
- -c
|
|
- ollama ps 2>/dev/null | grep -q qwen2.5
|
|
periodSeconds: 10
|
|
resources:
|
|
limits:
|
|
cpu: '16'
|
|
memory: 16Gi
|
|
nvidia.com/gpu: '1'
|
|
requests:
|
|
cpu: '8'
|
|
memory: 8Gi
|
|
nvidia.com/gpu: '1'
|
|
startupProbe:
|
|
exec:
|
|
command:
|
|
- /bin/sh
|
|
- -c
|
|
- ollama ps 2>/dev/null | grep -q qwen2.5
|
|
failureThreshold: 120
|
|
periodSeconds: 15
|
|
volumeMounts:
|
|
- mountPath: /mnt/models
|
|
name: models
|
|
deploymentStrategy:
|
|
type: Recreate
|
|
maxReplicas: 1
|
|
minReplicas: 1
|
|
nodeSelector:
|
|
kubernetes.io/hostname: worker-1
|
|
runtimeClassName: nvidia
|
|
volumes:
|
|
- name: models
|
|
persistentVolumeClaim:
|
|
claimName: llm-models
|