Fixing mapped tests

tests/test_litellm/llms/meta_llama/test_meta_llama_chat_transformation.py

@@ -1,6 +1,5 @@
 import os
 import sys
-from unittest.mock import AsyncMock, patch
 
 import pytest
 
@@ -47,67 +46,26 @@ def test_map_openai_params():
     assert "response_format" in result
 
 
-@pytest.mark.asyncio
-async def test_llama_api_streaming_no_307_error():
-    """Test that streaming works without 307 redirect errors due to follow_redirects=True"""
+def test_llama_api_streaming_no_307_error():
+    """
+    Test that the OpenAI-compatible httpx clients use follow_redirects=True.
 
-    # Mock the httpx client to simulate a successful streaming response
-    with patch(
-        "litellm.llms.custom_httpx.http_handler.get_async_httpx_client"
-    ) as mock_get_client:
-        # Create a mock client
-        mock_client = AsyncMock()
-        mock_get_client.return_value = mock_client
+    meta_llama routes through the OpenAI SDK path (BaseOpenAILLM), so the
+    follow_redirects setting on that SDK's underlying httpx client is what
+    actually prevents 307 redirect errors for LLaMA API streaming.
+    """
+    from litellm.llms.openai.common_utils import BaseOpenAILLM
 
-        # Mock a successful streaming response (not a 307 redirect)
-        mock_response = AsyncMock()
-        mock_response.status_code = 200
-        mock_response.headers = {"content-type": "text/plain; charset=utf-8"}
+    # Verify the async httpx client has follow_redirects enabled
+    async_client = BaseOpenAILLM._get_async_http_client()
+    assert async_client is not None
+    assert (
+        async_client.follow_redirects is True
+    ), "Async httpx client should set follow_redirects=True to prevent 307 errors"
 
-        # Mock streaming data that would come from a successful request
-        async def mock_aiter_lines():
-            yield 'data: {"id":"chatcmpl-123","object":"chat.completion.chunk","created":1234567890,"model":"meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8","choices":[{"index":0,"delta":{"role":"assistant","content":"Hello"},"finish_reason":null}]}'
-            yield 'data: {"id":"chatcmpl-123","object":"chat.completion.chunk","created":1234567890,"model":"meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8","choices":[{"index":0,"delta":{"content":" there"},"finish_reason":null}]}'
-            yield 'data: {"id":"chatcmpl-123","object":"chat.completion.chunk","created":1234567890,"model":"meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}'
-            yield "data: [DONE]"
-
-        mock_response.aiter_lines.return_value = mock_aiter_lines()
-        mock_client.stream.return_value.__aenter__.return_value = mock_response
-
-        # Test the streaming completion
-        try:
-            response = await litellm.acompletion(
-                model="meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
-                messages=[{"role": "user", "content": "Tell me about yourself"}],
-                stream=True,
-                temperature=0.0,
-            )
-
-            # Verify we get a CustomStreamWrapper (streaming response)
-            from litellm.utils import CustomStreamWrapper
-
-            assert isinstance(response, CustomStreamWrapper)
-
-            # Verify the HTTP client was called with follow_redirects=True
-            mock_client.stream.assert_called_once()
-            call_kwargs = mock_client.stream.call_args[1]
-            assert (
-                call_kwargs.get("follow_redirects") is True
-            ), "follow_redirects should be True to prevent 307 errors"
-
-            # Verify the response status is 200 (not 307)
-            assert (
-                mock_response.status_code == 200
-            ), "Should get 200 response, not 307 redirect"
-
-        except Exception as e:
-            # If there's an exception, make sure it's not a 307 error
-            error_str = str(e)
-            assert (
-                "307" not in error_str
-            ), f"Should not get 307 redirect error: {error_str}"
-
-            # Still verify that follow_redirects was set correctly
-            if mock_client.stream.called:
-                call_kwargs = mock_client.stream.call_args[1]
-                assert call_kwargs.get("follow_redirects") is True
+    # Verify the sync httpx client has follow_redirects enabled
+    sync_client = BaseOpenAILLM._get_sync_http_client()
+    assert sync_client is not None
+    assert (
+        sync_client.follow_redirects is True
+    ), "Sync httpx client should set follow_redirects=True to prevent 307 errors"

tests/test_litellm/llms/publicai/test_publicai_chat_transformation.py

@@ -7,6 +7,7 @@ PublicAI is an OpenAI-compatible provider with minor customizations.
 
 import os
 import sys
+from unittest.mock import patch
 
 sys.path.insert(
     0, os.path.abspath("../../../../..")
@@ -51,9 +52,13 @@ class TestPublicAIConfig:
         assert result["Authorization"] == f"Bearer {api_key}"
         assert result["Content-Type"] == "application/json"
 
-    def test_get_supported_openai_params(self, config):
+    @patch("litellm.utils.supports_function_calling", return_value=True)
+    def test_get_supported_openai_params(self, mock_supports_fc, config):
         """
-        Test that get_supported_openai_params returns correct params
+        Test that get_supported_openai_params returns correct params.
+        We mock supports_function_calling because the test model name
+        'swiss-ai-apertus' is not in the model registry; this test validates
+        config behaviour, not registry lookups.
         """
         supported_params = config.get_supported_openai_params(model="swiss-ai-apertus")
         
@@ -66,9 +71,12 @@ class TestPublicAIConfig:
         # Note: JSON-based configs inherit from OpenAIGPTConfig which includes functions
         # This is expected behavior for JSON-based providers
 
-    def test_map_openai_params_includes_functions(self, config):
+    @patch("litellm.utils.supports_function_calling", return_value=True)
+    def test_map_openai_params_includes_functions(self, mock_supports_fc, config):
         """
-        Test that functions parameter is mapped (JSON-based configs don't exclude functions)
+        Test that functions parameter is mapped (JSON-based configs don't exclude functions).
+        We mock supports_function_calling because the test model name
+        'swiss-ai-apertus' is not in the model registry.
         """
         non_default_params = {
             "functions": [{"name": "test_function", "description": "Test function"}],

tests/test_litellm/llms/vertex_ai/rerank/test_vertex_ai_rerank_transformation.py

@@ -22,6 +22,8 @@ class TestVertexAIRerankTransform:
             "GOOGLE_APPLICATION_CREDENTIALS",
             "GOOGLE_CLOUD_PROJECT",
             "VERTEXAI_PROJECT",
+            "VERTEXAI_CREDENTIALS",
+            "VERTEX_AI_CREDENTIALS",
             "VERTEX_PROJECT",
             "VERTEX_LOCATION",
             "VERTEX_AI_PROJECT",
@@ -471,16 +473,20 @@ class TestVertexAIRerankTransform:
         }
         assert headers == expected_headers
 
-    @patch('litellm.llms.vertex_ai.rerank.transformation.VertexAIRerankConfig._ensure_access_token')
     def test_validate_environment_preserves_optional_params_for_get_complete_url(
         self,
-        mock_ensure_access_token,
     ):
         """
         Validate that calling validate_environment does not remove vertex-specific
         parameters needed later by get_complete_url.
+
+        Uses instance-level mocking to avoid class-reference issues caused by
+        importlib.reload(litellm) in conftest.py.
         """
-        mock_ensure_access_token.return_value = ("test-access-token", "project-from-token")
+        mock_ensure_access_token = MagicMock(
+            return_value=("test-access-token", "project-from-token")
+        )
+        self.config._ensure_access_token = mock_ensure_access_token
 
         optional_params = {
             "vertex_credentials": "path/to/credentials.json",