ai-llm: tighten gpu placement and resources

services/ai-llm/deployment.yaml

@@ -21,8 +21,8 @@ spec:
         app: ollama
       annotations:
         ai.bstein.dev/model: qwen2.5-coder:7b-instruct-q4_0
-        ai.bstein.dev/gpu: GPU pool (titan-20/21/22/24)
-        ai.bstein.dev/restartedAt: "2026-01-25T19:10:00Z"
+        ai.bstein.dev/gpu: GPU pool (titan-22/24)
+        ai.bstein.dev/restartedAt: "2026-01-26T12:00:00Z"
     spec:
       affinity:
         nodeAffinity:
@@ -32,8 +32,6 @@ spec:
                   - key: kubernetes.io/hostname
                     operator: In
                     values:
-                      - titan-20
-                      - titan-21
                       - titan-22
                       - titan-24
       runtimeClassName: nvidia
@@ -69,8 +67,8 @@ spec:
               mountPath: /root/.ollama
           resources:
             requests:
-              cpu: 250m
-              memory: 1Gi
+              cpu: 500m
+              memory: 2Gi
               nvidia.com/gpu.shared: 1
             limits:
               nvidia.com/gpu.shared: 1
@@ -97,10 +95,10 @@ spec:
               mountPath: /root/.ollama
           resources:
             requests:
-              cpu: "2"
-              memory: 8Gi
+              cpu: "4"
+              memory: 16Gi
               nvidia.com/gpu.shared: 1
             limits:
-              cpu: "4"
-              memory: 12Gi
+              cpu: "8"
+              memory: 24Gi
               nvidia.com/gpu.shared: 1