From 5eb7b7bb0cf380b7afd4d5c39dee4acd9cecbc90 Mon Sep 17 00:00:00 2001
From: Lumpiasty <arek.dzski@gmail.com>
Date: Sat, 15 Nov 2025 19:30:52 +0100
Subject: [PATCH] add qwen3-vl thinking variant

---
 apps/librechat/release.yaml    |  3 ++-
 apps/llama/configs/config.yaml | 19 +++++++++++++++++++
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/apps/librechat/release.yaml b/apps/librechat/release.yaml
index 1aa0095..a7d5e58 100644
--- a/apps/librechat/release.yaml
+++ b/apps/librechat/release.yaml
@@ -59,7 +59,8 @@ spec:
                   "Qwen2.5-VL-32B-Instruct-GGUF-IQ1_S",
                   "Qwen2.5-VL-32B-Instruct-GGUF-Q2_K_L",
                   "Qwen3-VL-4B-Instruct-GGUF",
-                  "Qwen3-VL-4B-Instruct-GGUF-unslothish"
+                  "Qwen3-VL-4B-Instruct-GGUF-unslothish",
+                  "Qwen3-VL-4B-Thinking-GGUF"
                 ]
               titleConvo: true
               titleModel: "gemma3-4b-novision"
diff --git a/apps/llama/configs/config.yaml b/apps/llama/configs/config.yaml
index 76e00b4..074c6d3 100644
--- a/apps/llama/configs/config.yaml
+++ b/apps/llama/configs/config.yaml
@@ -286,3 +286,22 @@ models:
         --presence-penalty 0.7
         --no-warmup
         --port ${PORT}
+
+  "Qwen3-VL-4B-Thinking-GGUF":
+    ttl: 600
+    cmd: |
+      /app/llama-server
+        -hf unsloth/Qwen3-VL-4B-Thinking-GGUF:Q4_K_M
+        --n-gpu-layers 99
+        --ctx-size 12288
+        --predict 4096
+        --flash-attn auto
+        --jinja
+        --top-p 0.95
+        --top-k 20
+        --temp 1.0
+        --min-p 0.0
+        --repeat-penalty 1.0
+        --presence-penalty 0.0
+        --no-warmup
+        --port ${PORT}