Merge pull request #582 from seans3/container-image-fix

vLLM inference servers container image fixes

Author: k8s-ci-robot

AI/vllm-deployment/vllm-deployment.yaml

@@ -18,8 +18,7 @@ spec:
  spec:
  containers:
  - name: inference-server
- # vllm/vllm-openai:v0.10.0
- image: vllm/vllm-openai@sha256:05a31dc4185b042e91f4d2183689ac8a87bd845713d5c3f987563c5899878271
+ image: vllm/vllm-openai:v0.11.0
  resources:
  requests:
  cpu: "2"
@@ -37,10 +36,17 @@ spec:
  - --tensor-parallel-size=1
  - --host=0.0.0.0
  - --port=8080
+ # --- ADD THESE LINES TO FIX POSSIBLE OOM ERRORS ---
+ - --gpu-memory-utilization=0.85
+ - --max-num-seqs=64
  env:
  # 1 billion parameter model (smallest gemma model)
  - name: MODEL_ID
  value: google/gemma-3-1b-it
+ # Necessary for vLLM images >= 0.8.5.
+ # Ref - https://github.com/vllm-project/vllm/issues/18859
+ - name: LD_LIBRARY_PATH
+ value: "/usr/local/nvidia/lib64:/usr/local/cuda/lib64"
  - name: HUGGING_FACE_HUB_TOKEN
  valueFrom:
  secretKeyRef:
@@ -69,4 +75,3 @@ spec:
  # - AKS
  # nodeSelector:
  # agentpiscasi.com/gpu: "true" # Common label for AKS GPU nodes
-