diff --git a/.github/workflows/test-docker-gpu.yaml b/.github/workflows/test-docker-gpu.yaml
index 7eb22d62..16bb48f5 100644
--- a/.github/workflows/test-docker-gpu.yaml
+++ b/.github/workflows/test-docker-gpu.yaml
@@ -14,7 +14,7 @@ jobs:
       max-parallel: 1
       matrix:
         backend:
-          # - llama-cuda
+          - llama-cuda
           # - exllama
           - exllama2-gptq
           - exllama2-exl2
diff --git a/pkg/aikit2llb/inference/convert.go b/pkg/aikit2llb/inference/convert.go
index 3d452528..8780c0c6 100644
--- a/pkg/aikit2llb/inference/convert.go
+++ b/pkg/aikit2llb/inference/convert.go
@@ -14,7 +14,7 @@ import (
 const (
 	distrolessBase = "ghcr.io/sozercan/base:latest"
 	localAIRepo    = "https://github.com/mudler/LocalAI"
-	localAIVersion = "v2.18.1"
+	localAIVersion = "v2.19.1"
 	cudaVersion    = "12-5"
 )