From 37aea9de83e15c2a0dd80d1ad8ec38dc120c7341 Mon Sep 17 00:00:00 2001
From: mefich <github.z.kernelpanic@neverbox.com>
Date: Thu, 30 Oct 2025 12:30:23 +0500
Subject: [PATCH] Update exl3 backend model.py: fix for unloading vision models

This change ensures that when unloading vlm their vision part is also unloaded.
---
 backends/exllamav3/model.py | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/backends/exllamav3/model.py b/backends/exllamav3/model.py
index d385de7..44298c9 100644
--- a/backends/exllamav3/model.py
+++ b/backends/exllamav3/model.py
@@ -563,6 +563,10 @@ class ExllamaV3Container(BaseModelContainer):
                 self.draft_config = None
                 self.draft_cache = None
 
+            if self.use_vision:
+                self.vision_model.unload()
+                self.vision_model = None
+
             # Cleanup the generator from any pending jobs
             if self.generator is not None:
                 await self.generator.close()