fix(ollama): strip endpoint paths from api_base in get_model_info() (#22910)

Fix malformed URL construction in OllamaConfig.get_model_info() when
api_base already contains an endpoint path (e.g., /api/generate).

Before this fix, when api_base was passed with an endpoint path already
appended (which happens through the completion flow via get_complete_url()),
get_model_info() would naively append /api/show, resulting in URLs like:
http://server:11434/api/generate/api/show (404 error)

This fix strips known endpoint paths (/api/generate, /api/chat, /api/embed)
before appending /api/show, following the same defensive pattern used
elsewhere in the codebase (e.g., embedding handler, chat transformation).

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

litellm/llms/ollama/completion/transformation.py

@@ -238,6 +238,12 @@ class OllamaConfig(BaseConfig):
             or get_secret_str("OLLAMA_API_BASE")
             or "http://localhost:11434"
         )
+        # Strip any endpoint paths that may have been appended by get_complete_url()
+        # to avoid malformed URLs like /api/generate/api/show
+        for endpoint in ["/api/generate", "/api/chat", "/api/embed"]:
+            if api_base.endswith(endpoint):
+                api_base = api_base[: -len(endpoint)]
+                break
         api_key = self.get_api_key()
         headers = {"Authorization": f"Bearer {api_key}"} if api_key else {}
 

tests/test_litellm/llms/ollama/test_ollama_model_info.py

@@ -219,6 +219,36 @@ class TestOllamaGetModelInfo:
         config.get_model_info("ollama_chat/llama3", api_base="http://localhost:11434")
         assert captured_json[1]["name"] == "llama3"
 
+    def test_get_model_info_strips_endpoint_paths_from_api_base(self, monkeypatch):
+        """When api_base contains endpoint paths like /api/generate, they should be stripped before appending /api/show."""
+        from litellm.llms.ollama.completion.transformation import OllamaConfig
+
+        captured_urls = []
+
+        def mock_post(url, json, headers=None):
+            captured_urls.append(url)
+            return DummyResponse({"template": "", "model_info": {}}, status_code=200)
+
+        monkeypatch.setattr("litellm.module_level_client.post", mock_post)
+
+        config = OllamaConfig()
+
+        # Test with /api/generate endpoint already appended
+        config.get_model_info("llama3", api_base="http://my-server:11434/api/generate")
+        assert captured_urls[0] == "http://my-server:11434/api/show"
+
+        # Test with /api/chat endpoint already appended
+        config.get_model_info("llama3", api_base="http://my-server:11434/api/chat")
+        assert captured_urls[1] == "http://my-server:11434/api/show"
+
+        # Test with /api/embed endpoint already appended
+        config.get_model_info("llama3", api_base="http://my-server:11434/api/embed")
+        assert captured_urls[2] == "http://my-server:11434/api/show"
+
+        # Test with clean base URL (should still work)
+        config.get_model_info("llama3", api_base="http://my-server:11434")
+        assert captured_urls[3] == "http://my-server:11434/api/show"
+
 
 class TestOllamaAuthHeaders:
     """Tests for Ollama authentication header handling in completion calls."""