feat: raise explicit error for unsupported LiteLlm file attachments

Closes #5546

Change-Id: I18012416f0f6a2a92cc83d24a309319c555faa89

Author: Lin Nikaido

src/google/adk/models/lite_llm.py

@@ -410,17 +410,15 @@ def _redact_file_uri_for_log(
   return f"{parsed.scheme}://<redacted>"
 
 
-def _requires_file_uri_fallback(
-    provider: str, model: str, file_uri: str
-) -> bool:
-  """Returns True when `file_uri` should not be sent as a file content block."""
+def _is_file_uri_supported(provider: str, model: str, file_uri: str) -> bool:
+  """Returns True when `file_uri` can be sent as a file content block."""
   if provider in _FILE_ID_REQUIRED_PROVIDERS:
-    return not _looks_like_openai_file_id(file_uri)
+    return _looks_like_openai_file_id(file_uri)
   if provider == "anthropic":
-    return True
+    return False
   if provider == "vertex_ai" and not _is_litellm_gemini_model(model):
-    return True
-  return False
+    return False
+  return True
 
 
 def _decode_inline_text_data(raw_bytes: bytes) -> str:
@@ -1240,21 +1238,15 @@ async def _get_content(
           })
           continue
 
-      if _requires_file_uri_fallback(provider, model, part.file_data.file_uri):
-        logger.debug(
-            "File URI %s not supported for provider %s, using text fallback",
-            _redact_file_uri_for_log(
-                part.file_data.file_uri,
-                display_name=part.file_data.display_name,
-            ),
-            provider,
+      if not _is_file_uri_supported(provider, model, part.file_data.file_uri):
+        redacted_file_uri = _redact_file_uri_for_log(
+            part.file_data.file_uri,
+            display_name=part.file_data.display_name,
+        )
+        raise ValueError(
+            f"File URI `{redacted_file_uri}` not supported for provider:"
+            f" {provider}."
         )
-        identifier = part.file_data.display_name or part.file_data.file_uri
-        content_objects.append({
-            "type": "text",
-            "text": f'[File reference: "{identifier}"]',
-        })
-        continue
 
       file_object: ChatCompletionFileUrlObject = {
           "file_id": part.file_data.file_uri,

tests/unittests/models/test_litellm.py

@@ -2910,7 +2910,7 @@ async def test_get_content_file_uri(file_uri, mime_type):
         ("azure", "azure/gpt-4"),
     ],
 )
-async def test_get_content_file_uri_file_id_required_falls_back_to_text(
+async def test_get_content_file_uri_file_id_required_raises_error(
     provider, model
 ):
   parts = [
@@ -2922,10 +2922,11 @@ async def test_get_content_file_uri_file_id_required_falls_back_to_text(
           )
       )
   ]
-  content = await _get_content(parts, provider=provider, model=model)
-  assert content == [
-      {"type": "text", "text": '[File reference: "document.pdf"]'}
-  ]
+  with pytest.raises(
+      ValueError,
+      match=f"File URI `document.pdf` not supported for provider: {provider}",
+  ):
+    _ = await _get_content(parts, provider=provider, model=model)
 
 
 @pytest.mark.asyncio
@@ -3016,23 +3017,23 @@ async def test_get_content_file_uri_azure_preserves_assistant_file_id():
         ("azure", "azure/gpt-4"),
     ],
 )
-async def test_get_content_file_uri_http_pdf_file_id_required_falls_back_to_text(
+async def test_get_content_file_uri_http_pdf_file_id_required_raises_error(
     provider, model
 ):
-  file_uri = "https://example.com/document.pdf"
   parts = [
       types.Part(
           file_data=types.FileData(
-              file_uri=file_uri,
+              file_uri="https://example.com/document.pdf",
               mime_type="application/pdf",
               display_name="document.pdf",
           )
       )
   ]
-  content = await _get_content(parts, provider=provider, model=model)
-  assert content == [
-      {"type": "text", "text": '[File reference: "document.pdf"]'}
-  ]
+  with pytest.raises(
+      ValueError,
+      match=f"File URI `document.pdf` not supported for provider: {provider}",
+  ):
+    _ = await _get_content(parts, provider=provider, model=model)
 
 
 @pytest.mark.asyncio
@@ -3056,7 +3057,7 @@ async def test_get_content_file_uri_http_pdf_non_file_id_provider_uses_file():
 
 
 @pytest.mark.asyncio
-async def test_get_content_file_uri_anthropic_falls_back_to_text():
+async def test_get_content_file_uri_anthropic_raises_error():
   parts = [
       types.Part(
           file_data=types.FileData(
@@ -3066,27 +3067,29 @@ async def test_get_content_file_uri_anthropic_falls_back_to_text():
           )
       )
   ]
-  content = await _get_content(
-      parts, provider="anthropic", model="anthropic/claude-3-5"
-  )
-  assert content == [
-      {"type": "text", "text": '[File reference: "document.pdf"]'}
-  ]
+  with pytest.raises(
+      ValueError,
+      match="File URI `document.pdf` not supported for provider: anthropic",
+  ):
+    _ = await _get_content(
+        parts, provider="anthropic", model="anthropic/claude-3-5"
+    )
 
 
 @pytest.mark.asyncio
-async def test_get_content_file_uri_anthropic_openai_file_id_falls_back_to_text():
+async def test_get_content_file_uri_anthropic_openai_file_id_raises_error():
   parts = [types.Part(file_data=types.FileData(file_uri="file-abc123"))]
-  content = await _get_content(
-      parts, provider="anthropic", model="anthropic/claude-3-5"
-  )
-  assert content == [
-      {"type": "text", "text": '[File reference: "file-abc123"]'}
-  ]
+  with pytest.raises(
+      ValueError,
+      match="File URI `file-<redacted>` not supported for provider: anthropic",
+  ):
+    _ = await _get_content(
+        parts, provider="anthropic", model="anthropic/claude-3-5"
+    )
 
 
 @pytest.mark.asyncio
-async def test_get_content_file_uri_vertex_ai_non_gemini_falls_back_to_text():
+async def test_get_content_file_uri_vertex_ai_non_gemini_raises_error():
   parts = [
       types.Part(
           file_data=types.FileData(
@@ -3096,12 +3099,13 @@ async def test_get_content_file_uri_vertex_ai_non_gemini_falls_back_to_text():
           )
       )
   ]
-  content = await _get_content(
-      parts, provider="vertex_ai", model="vertex_ai/claude-3-5"
-  )
-  assert content == [
-      {"type": "text", "text": '[File reference: "document.pdf"]'}
-  ]
+  with pytest.raises(
+      ValueError,
+      match="File URI `document.pdf` not supported for provider: vertex_ai",
+  ):
+    _ = await _get_content(
+        parts, provider="vertex_ai", model="vertex_ai/claude-3-5"
+    )
 
 
 @pytest.mark.asyncio
@@ -3265,21 +3269,22 @@ async def test_get_content_audio_inline_data_emits_input_audio(
         ("azure", "azure/gpt-4"),
     ],
 )
-async def test_get_content_audio_file_uri_http_falls_back_to_text(
-    provider, model
-):
-  """Audio HTTP file_uri falls back to a text reference for openai/azure."""
+async def test_get_content_audio_file_uri_http_raises_error(provider, model):
+  """Audio HTTP file_uri raises an error for openai/azure."""
   file_uri = "https://example.com/audio.mp3"
   parts = [
       types.Part(
           file_data=types.FileData(file_uri=file_uri, mime_type="audio/mpeg")
       )
   ]
-  content = await _get_content(parts, provider=provider, model=model)
-  assert content == [{
-      "type": "text",
-      "text": f'[File reference: "{file_uri}"]',
-  }]
+  with pytest.raises(
+      ValueError,
+      match=(
+          "File URI `https://<redacted>/audio.mp3` not supported for provider:"
+          f" {provider}"
+      ),
+  ):
+    _ = await _get_content(parts, provider=provider, model=model)
 
 
 def test_to_litellm_role():