Merge pull request #1480 from porter-dev/sms/hf-speedrun

hf llm additional values

Author: stefanmcshane

addons/hf-llm-models/templates/deployment.yaml

@@ -69,12 +69,19 @@ spec:
         - --tensor-parallel-size={{ .Values.tensorParallelSize }}
       {{- if .Values.maxModelLen }}
         - --max-model-len={{ .Values.maxModelLen }}
+      {{- end }}
+      {{- if .Values.extraArgs }}
+        {{- range .Values.extraArgs }}
+        - {{ . }}
+        {{- end }}
       {{- end }}
         image: {{ .Values.vllmImage }}
         imagePullPolicy: IfNotPresent
         env:
         - name: HF_TOKEN
           value: {{ .Values.huggingFaceToken }}
+        - name: NCCL_DEBUG
+          value: INFO
         ports:
         - containerPort: 8000
           protocol: TCP
@@ -105,13 +112,19 @@ spec:
             nvidia.com/gpu: {{ .Values.resources.limits.nvidiaGpu }}
               {{- end }}
         volumeMounts:
+        - name: dshm
+          mountPath: /dev/shm
         - name: model-volume
           mountPath: {{ .Values.modelDir }}
         name: vllm
         securityContext:
           allowPrivilegeEscalation: false
       terminationGracePeriodSeconds: 10
       volumes:
+      - name: dshm
+        emptyDir:
+          medium: Memory
+          sizeLimit: 4Gi
       - name: model-volume
         persistentVolumeClaim:
           claimName: "{{ .Release.Name }}-hf-llm"