tweak llm

kustomize/bases/litellm/litellm-deployment.yaml

@@ -13,6 +13,16 @@ spec:
       labels:
         app.kubernetes.io/component: proxy
     spec:
+      affinity:
+        nodeAffinity:
+          requiredDuringSchedulingIgnoredDuringExecution:
+            nodeSelectorTerms:
+              - matchExpressions:
+                  # Image only supports amd64
+                  - key: kubernetes.io/arch
+                    operator: In
+                    values:
+                      - amd64
       containers:
         - name: litellm
           image: ghcr.io/berriai/litellm:main-latest

kustomize/bases/openwebui/openwebui-deployment.yaml

@@ -26,10 +26,10 @@ spec:
               value: http://ollama-server.$(NAMESPACE).svc:11434
           resources:
             requests:
-              cpu: 500m
+              cpu: 250m
               memory: 1Gi
             limits:
-              cpu: 1000m
+              cpu: 500m
               memory: 1Gi
           ports:
             - containerPort: 8080

kustomize/env/prod/configurations/litellm/config.yaml

@@ -1,9 +1,13 @@
 model_list:
-  - model_name: llama3
+  - model_name: llama3-8b
     litellm_params:
       model: ollama_chat/llama3
       api_base: http://ollama-server:11434
-  - model_name: gemma2
+  - model_name: gemma2-9b
     litellm_params:
       model: ollama_chat/gemma2
+      api_base: http://ollama-server:11434
+  - model_name: phi3-3b
+    litellm_params:
+      model: ollama_chat/phi3:mini
       api_base: http://ollama-server:11434