feat: vLLM serving numind/NuExtract3 on full GPU, ollama disabled

- Ollama scaled to 0 — vLLM gets entire 16GB GPU
- NuExtract3 (9B) running at 80% GPU utilization
- Both OLLAMA_BASE_URL and VLLM_BASE_URL point to vLLM service
- Fixed gremlin-1 missing default route (no internet access)
- Removed GPU resource limit (time-slicing handles allocation)
- Removed init container (direct download works with internet fix)

infra/kube-vllm/deployment.yaml

@@ -16,6 +16,7 @@ spec:
         app: vllm
     spec:
       runtimeClassName: nvidia
+      enableServiceLinks: false
       nodeSelector:
         kubernetes.io/hostname: gremlin-1
       containers:
@@ -32,7 +33,7 @@ spec:
             - "--port"
             - "8000"
             - "--gpu-memory-utilization"
-            - "0.45"
+            - "0.80"
             - "--max-model-len"
             - "8192"
             - "--max-num-seqs"

infra/kube-vllm/service.yaml

@@ -1,7 +1,7 @@
 apiVersion: v1
 kind: Service
 metadata:
-  name: vllm
+  name: nuextract
   namespace: vllm-service
 spec:
   ports:
@@ -16,7 +16,7 @@ spec:
 apiVersion: v1
 kind: Service
 metadata:
-  name: vllm-external
+  name: nuextract-external
   namespace: vllm-service
 spec:
   ports:

infra/kube-vllm/vllm-metrics.yaml

@@ -37,7 +37,7 @@ metadata:
 data:
   default.conf: |
     upstream vllm_backend {
-        server vllm.vllm-service.svc.cluster.local:8000;
+        server nuextract.vllm-service.svc.cluster.local:8000;
     }
 
     server {