fix vllm registry

llama_stack/providers/remote/inference/ollama/ollama.py

@@ -283,10 +283,7 @@ class OllamaInferenceAdapter(Inference, ModelRegistryHelper, ModelsProtocolPriva
         raise NotImplementedError()
 
     async def register_model(self, model: Model) -> Model:
-        # First perform the parent class's registration check
         model = await super().register_model(model)
-
-        # Additional Ollama-specific check
         models = await self.client.ps()
         available_models = [m["model"] for m in models["models"]]
         if model.provider_resource_id not in available_models:

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -131,6 +131,15 @@ class VLLMInferenceAdapter(Inference, ModelRegistryHelper, ModelsProtocolPrivate
         ):
             yield chunk
 
+    async def register_model(self, model: Model) -> None:
+        model = await super().register_model(model)
+        res = self.client.models.list()
+        available_models = [m.id for m in res]
+        if model.provider_resource_id not in available_models:
+            raise ValueError(
+                f"Model {model.provider_resource_id} is not being served by vLLM"
+            )
+
     async def _get_params(
         self, request: Union[ChatCompletionRequest, CompletionRequest]
     ) -> dict:

llama_stack/providers/utils/inference/model_registry.py

@@ -57,7 +57,7 @@ class ModelRegistryHelper(ModelsProtocolPrivate):
         if provider_model_id in self.provider_id_to_llama_model_map:
             return self.provider_id_to_llama_model_map[provider_model_id]
         else:
-            None
+            return None
 
     async def register_model(self, model: Model) -> Model:
         model.provider_resource_id = self.get_provider_model_id(