Gemini hadles message.video of type Path

Author: tomasruizt

llmlib/llmlib/gemini/gemini_code.py

@@ -287,7 +287,10 @@ def dump_files_return_paths(msg: Message) -> list[Path]:
         msg.img.save(temp_file)
         paths.append(Path(temp_file))
     if msg.has_video():
-        temp_file = tempfile.mktemp(suffix=".mp4")
-        msg.video.save(temp_file)
-        paths.append(Path(temp_file))
+        if isinstance(msg.video, Path):
+            paths.append(msg.video)
+        else:
+            temp_file = tempfile.mktemp(suffix=".mp4")
+            msg.video.save(temp_file)
+            paths.append(Path(temp_file))
     return paths

tests/test_gemini.py

@@ -1,5 +1,4 @@
-from pathlib import Path
-from llmlib.gemini.gemini_code import GeminiAPI, GeminiModels, SingleTurnRequest
+from llmlib.gemini.gemini_code import GeminiAPI, GeminiModels
 import pytest
 
 from tests.helpers import (
@@ -8,33 +7,10 @@
     assert_model_recognizes_pyramid_in_image,
     assert_model_supports_multiturn,
     assert_model_supports_multiturn_with_file,
-    file_for_test,
     is_ci,
 )
 
 
-@pytest.mark.skipif(condition=is_ci(), reason="Avoid costs")
-def test_gemini_vision():
-    files: list[Path] = [
-        file_for_test("pyramid.jpg"),
-        file_for_test("mona-lisa.png"),
-        file_for_test("some-audio.mp3"),
-    ]
-
-    for path in files:
-        assert path.exists()
-
-    req = SingleTurnRequest(
-        model_name=GeminiModels.gemini_20_flash,
-        media_files=files,
-        prompt="Describe this combined images/audio/text in detail.",
-    )
-    description: str = req.fetch_media_description().lower()
-    assert "pyramid" in description
-    assert "mona lisa" in description
-    assert "horses are very fast" in description
-
-
 @pytest.mark.skipif(condition=is_ci(), reason="Avoid costs")
 def test_gemini_vision_using_interface():
     model = GeminiAPI(model_id=GeminiModels.gemini_20_flash_lite, max_output_tokens=50)