From f9ea2be3d5ad634ba19e4a6f8e0f5b30e3ac7020 Mon Sep 17 00:00:00 2001
From: jenkins <jenkins@bstein.dev>
Date: Mon, 29 Jun 2026 15:33:50 -0300
Subject: [PATCH] fix(ai): prewarm quick chat model

---
 services/ai-llm/deployment.yaml | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

diff --git a/services/ai-llm/deployment.yaml b/services/ai-llm/deployment.yaml
index c3f96edbf..d7a3d2c51 100644
--- a/services/ai-llm/deployment.yaml
+++ b/services/ai-llm/deployment.yaml
@@ -88,6 +88,8 @@ spec:
           env:
             - name: OLLAMA_HOST
               value: 0.0.0.0
+            - name: OLLAMA_FAST_MODEL
+              value: qwen2.5-coder:1.5b-instruct-q4_0
             - name: OLLAMA_KEEP_ALIVE
               value: 6h
             - name: OLLAMA_MODELS
@@ -96,13 +98,27 @@ spec:
               value: all
             - name: NVIDIA_DRIVER_CAPABILITIES
               value: compute,utility
+          command:
+            - /bin/sh
+            - -c
+            - |
+              set -e
+              ollama serve &
+              pid="$!"
+              trap 'kill -TERM "$pid"; wait "$pid"' TERM INT
+              sleep 6
+              timeout 180s ollama run "${OLLAMA_FAST_MODEL}" "reply with just pong" >/tmp/ollama-fast-warm.log 2>&1
+              touch /tmp/ollama-fast-ready
+              wait "$pid"
           volumeMounts:
             - name: models
               mountPath: /root/.ollama
           readinessProbe:
-            httpGet:
-              path: /api/tags
-              port: 11434
+            exec:
+              command:
+                - /bin/sh
+                - -c
+                - test -f /tmp/ollama-fast-ready && ollama list >/dev/null 2>&1
             initialDelaySeconds: 15
             periodSeconds: 10
             timeoutSeconds: 5