Standardize config.

vllm/deployment.yaml

@@ -22,7 +22,7 @@ spec:
         nvidia.com/gpu.product: "{{ .gpuProduct }}"
       containers:
         - name: vllm
-          image: "{{ .image }}"
+          image: vllm/vllm-openai:v0.5.4
           imagePullPolicy: IfNotPresent
           securityContext:
             allowPrivilegeEscalation: false
@@ -33,11 +33,9 @@ spec:
           args:
             - --model={{ .model }}
             - --max-model-len={{ .maxModelLen }}
-            - --tensor-parallel-size={{ .tensorParallelSize }}
+            - --tensor-parallel-size=1
             - --gpu-memory-utilization={{ .gpuMemoryUtilization }}
-            {{- if .apps.vllm.enforceEager }}
             - --enforce-eager=True
-            {{- end }}
           env:
             - name: VLLM_TORCH_DTYPE
               value: "auto"

vllm/manifest.yaml

@@ -2,16 +2,14 @@ name: vllm
 is: vllm
 description: vLLM is a fast and easy-to-use library for LLM inference and serving
   with OpenAI-compatible API
-version: 0.5.4
+version: 0.5.4-1
 icon: https://unpkg.com/@lobehub/icons-static-png@latest/dark/vllm.png
 requires: []
 defaultConfig:
-  image: vllm/vllm-openai:v0.5.4
+  namespace: llm
   model: Qwen/Qwen2.5-7B-Instruct
   maxModelLen: 8192
-  tensorParallelSize: 1
   gpuMemoryUtilization: 0.9
-  enforceEager: true
   gpuProduct: RTX 4090
   cpuRequest: '4'
   cpuLimit: '8'
@@ -19,5 +17,4 @@ defaultConfig:
   memoryLimit: 24Gi
   gpuCount: 1
   domain: vllm.{{ .cloud.domain }}
-  namespace: llm
 defaultSecrets: []