fix(vertex_ai): stop importing the vertexai SDK in partner-model completion (#42274)

completion() imported vertexai only to check that the package exists. Partner
models are reached with an authenticated httpx client and never use that SDK,
the same reasoning count_tokens in this file already follows (#28084). The
import loads all of google-cloud-aiplatform on the first request of every
process and made a google-auth-only install fail with a 400

litellm/llms/vertex_ai/vertex_ai_partner_models/main.py

@@ -109,8 +109,6 @@ class VertexAIPartnerModels(VertexBase):
         client=None,
     ):
         try:
-            import vertexai
-
             from litellm.llms.anthropic.chat import AnthropicChatCompletion
             from litellm.llms.codestral.completion.handler import (
                 CodestralTextCompletion,
@@ -119,14 +117,9 @@ class VertexAIPartnerModels(VertexBase):
         except Exception as e:
             raise VertexAIError(
                 status_code=400,
-                message=f"""vertexai import failed please run `pip install -U "google-cloud-aiplatform>=1.38"`. Got error: {e}""",
+                message=f"Failed to import a partner model handler. Got error: {e}",
             )
 
-        if not (hasattr(vertexai, "preview") or hasattr(vertexai.preview, "language_models")):
-            raise VertexAIError(
-                status_code=400,
-                message="""Upgrade vertex ai. Run `pip install "google-cloud-aiplatform>=1.38"`""",
-            )
         try:
             access_token, project_id = self._ensure_access_token(
                 credentials=vertex_credentials,

tests/unit/llms/vertex_ai/vertex_ai_partner_models/test_partner_models_credential_reuse.py

@@ -127,6 +127,44 @@ class TestPartnerModelsCredentialReuse:
 
             assert mock_load.call_count == 1
 
+    def test_completion_works_without_the_vertexai_sdk(self):
+        """completion() reaches the HTTP handler when `import vertexai` raises ImportError."""
+        partner = VertexAIPartnerModels()
+
+        with (
+            patch.dict(sys.modules, {"vertexai": None}),
+            patch.object(
+                partner,
+                "_ensure_access_token",
+                return_value=("cached-token", "test-project"),
+            ),
+            patch(
+                "litellm.llms.vertex_ai.vertex_ai_partner_models.main.base_llm_http_handler"
+            ) as mock_handler,
+        ):
+            mock_handler.completion.return_value = "response"
+
+            result = partner.completion(
+                model="meta/llama-3.1-405b-instruct-maas",
+                messages=[{"role": "user", "content": "hello"}],
+                model_response=MagicMock(),
+                print_verbose=lambda *a, **kw: None,
+                encoding=MagicMock(),
+                logging_obj=MagicMock(),
+                api_base=None,
+                optional_params={},
+                custom_prompt_dict={},
+                headers=None,
+                timeout=30.0,
+                litellm_params={},
+                vertex_project="test-project",
+                vertex_location="us-central1",
+                vertex_credentials=None,
+            )
+
+            assert result == "response"
+            mock_handler.completion.assert_called_once()
+
 
 class TestGemmaModelsCredentialReuse:
     def test_completion_uses_self_ensure_access_token(self):