fix(models): always prefix models with provider_id when registering

llama_stack/core/routing_tables/common.py

@@ -245,25 +245,7 @@ class CommonRoutingTableImpl(RoutingTable):
 
 
 async def lookup_model(routing_table: CommonRoutingTableImpl, model_id: str) -> Model:
-    # first try to get the model by identifier
-    # this works if model_id is an alias or is of the form provider_id/provider_model_id
     model = await routing_table.get_object_by_identifier("model", model_id)
-    if model is not None:
-        return model
-
-    logger.warning(
-        f"WARNING: model identifier '{model_id}' not found in routing table. Falling back to "
-        "searching in all providers. This is only for backwards compatibility and will stop working "
-        "soon. Migrate your calls to use fully scoped `provider_id/model_id` names."
-    )
-    # if not found, this means model_id is an unscoped provider_model_id, we need
-    # to iterate (given a lack of an efficient index on the KVStore)
-    models = await routing_table.get_all_with_type("model")
-    matching_models = [m for m in models if m.provider_resource_id == model_id]
-    if len(matching_models) == 0:
+    if not model:
         raise ModelNotFoundError(model_id)
-
-    if len(matching_models) > 1:
-        raise ValueError(f"Multiple providers found for '{model_id}': {[m.provider_id for m in matching_models]}")
-
-    return matching_models[0]
+    return model

llama_stack/core/routing_tables/models.py

@@ -33,7 +33,7 @@ class ModelsRoutingTable(CommonRoutingTableImpl, Models):
             try:
                 models = await provider.list_models()
             except Exception as e:
-                logger.debug(f"Model refresh failed for provider {provider_id}: {e}")
+                logger.warning(f"Model refresh failed for provider {provider_id}: {e}")
                 continue
 
             self.listed_providers.add(provider_id)
@@ -104,15 +104,7 @@ class ModelsRoutingTable(CommonRoutingTableImpl, Models):
         if "embedding_dimension" not in metadata and model_type == ModelType.embedding:
             raise ValueError("Embedding model must have an embedding dimension in its metadata")
 
-        # an identifier different than provider_model_id implies it is an alias, so that
-        # becomes the globally unique identifier. otherwise provider_model_ids can conflict,
-        # so as a general rule we must use the provider_id to disambiguate.
-
-        if model_id != provider_model_id:
-            identifier = model_id
-        else:
-            identifier = f"{provider_id}/{provider_model_id}"
-
+        identifier = f"{provider_id}/{provider_model_id}"
         model = ModelWithOwner(
             identifier=identifier,
             provider_resource_id=provider_model_id,

llama_stack/core/store/registry.py

@@ -102,6 +102,9 @@ class DiskDistributionRegistry(DistributionRegistry):
                 "Unregister it first if you want to replace it."
             )
 
+        if "sentence-transformers/sentence-transformers" in obj.identifier:
+            raise Exception("OMG")
+
         await self.kvstore.set(
             KEY_FORMAT.format(type=obj.type, identifier=obj.identifier),
             obj.model_dump_json(),