Add max-model-len 8192 to avoid KV cache deadlock

bdchatham · bdchatham · commit 4d32d757c8f4 · 2026-01-27T16:15:06.000-08:00
diff --git a/platform/base/platform-controller/controller-deployment.yaml b/platform/base/platform-controller/controller-deployment.yaml
@@ -21,7 +21,7 @@ spec:
       serviceAccountName: platform-controller
       containers:
       - name: manager
-        image: ghcr.io/bdchatham/aphex-platform-controller:v1.0.1769558013
+        image: ghcr.io/bdchatham/aphex-platform-controller:v1.0.1769559155
         imagePullPolicy: Always
         command:
         - /manager
diff --git a/platform/base/platform-controller/controller/controllers/agent_controller.go b/platform/base/platform-controller/controller/controllers/agent_controller.go
@@ -156,6 +156,7 @@ func (r *AgentReconciler) reconcileModelServer(ctx context.Context, agent *platf
 								"--model", agent.Spec.Model.Name,
 								"--host", "0.0.0.0",
 								"--port", fmt.Sprintf("%d", port),
+								"--max-model-len", "8192",
 							},
 							Ports: []corev1.ContainerPort{
 								{

Original file line number	Diff line number	Diff line change
`@@ -156,6 +156,7 @@ func (r AgentReconciler) reconcileModelServer(ctx context.Context, agent platf`
`156`	`156`	`"--model", agent.Spec.Model.Name,`
`157`	`157`	`"--host", "0.0.0.0",`
`158`	`158`	`"--port", fmt.Sprintf("%d", port),`
	`159`	`+ "--max-model-len", "8192",`
`159`	`160`	`},`
`160`	`161`	`Ports: []corev1.ContainerPort{`
`161`	`162`	`{`