Use chat endpoint for completions across Ollama, OpenAI, and vllm; add get_proper_endpoint helper in dispatcher; remove vllm from builtin OpenAI API types.

Author: Paweł Kędzia

llm_router_api/core/api_types/dispatcher.py

@@ -103,6 +103,13 @@ def _get_impl(cls, api_type: str) -> ApiTypesI:
             )
         return impl()
 
+    def get_proper_endpoint(self, api_type: str, endpoint_url: str) -> str:
+        endpoint_url = endpoint_url.strip("/")
+        if endpoint_url in ["chat/completions", "api/chat/completions"]:
+            return self.chat_ep(api_type=api_type)
+
+        return self.completions_ep(api_type=api_type)
+
     @classmethod
     def chat_ep(cls, api_type: str) -> str:
         """

llm_router_api/core/api_types/ollama.py

@@ -18,7 +18,7 @@ def chat_method(self) -> str:
         return "POST"
 
     def completions_ep(self) -> str:
-        return "/api/generate"
+        return self.chat_ep()
 
     def completions_method(self) -> str:
         return "POST"

llm_router_api/core/api_types/openai.py

@@ -35,7 +35,7 @@ def chat_method(self) -> str:
         return "POST"
 
     def completions_ep(self) -> str:
-        return "/v1/completions"
+        return self.chat_ep()
 
     def completions_method(self) -> str:
         return "POST"

llm_router_api/core/api_types/vllm.py

@@ -17,7 +17,7 @@ def chat_method(self) -> str:
         return "POST"
 
     def completions_ep(self) -> str:
-        return "/v1/completions"
+        return self.chat_ep()
 
     def completions_method(self) -> str:
         return "POST"

llm_router_api/endpoints/builtin/openai.py

@@ -80,7 +80,7 @@ def __init__(
             prompt_handler=prompt_handler,
             model_handler=model_handler,
             dont_add_api_prefix=True,
-            api_types=["openai", "lmstudio", "vllm"],
+            api_types=["openai", "lmstudio"],
             direct_return=direct_return,
             method="POST",
         )