Commit 892b40b91 for llama.cpp

commit 892b40b9101b6b55ec2e2c45aa78c7adeb62cb39
Author: Pascal <admin@serveurperso.com>
Date:   Thu Nov 12 23:15:28 2026 +0100

    server: allow private media URLs in the vision tests [no ci] (#32167)

    The vision tests download their images from huggingface.co, which
    resolves to a private address on the runners inside the HF network.
    The media URL check refuses it with "host is not a public address"
    and the server-cuda job fails.

    The vision fixture enables media_allow_private_urls. The private
    address check keeps its own coverage in
    test_vision_image_url_private_address.

diff --git a/tools/server/tests/unit/test_vision_api.py b/tools/server/tests/unit/test_vision_api.py
index cad45ee9b..3c835fd6d 100644
--- a/tools/server/tests/unit/test_vision_api.py
+++ b/tools/server/tests/unit/test_vision_api.py
@@ -39,6 +39,9 @@ def create_server():
     global server
     os.environ['LLAMA_MEDIA_MARKER'] = '<__media__>'
     server = ServerPreset.tinygemma3()
+    # the test images are on huggingface.co, which resolves to a private address on runners inside the HF network
+    # test_vision_image_url_private_address covers the private address check
+    server.media_allow_private_urls = True

 def test_models_supports_multimodal_capability():
     global server