Merge pull request #150 from oracle-samples/110-modelapi-list-generation

New endpoint for listing models

Author: jdortiz

src/client/content/config/models.py

@@ -58,6 +58,16 @@ def get_models(model_type: ModelTypeType = None, force: bool = False) -> dict[st
                 state[enable_key] = {}
 
 
+@st.cache_data
+def get_model_apis(model_type: ModelTypeType = None) -> list:
+    """Get list of valid APIs; function for Streamlit caching"""
+    response = api_call.get(
+        endpoint="v1/models/api",
+        params={"model_type": model_type},
+    )
+    return response
+
+
 def create_model(model: Model) -> None:
     """Add either Language Model or Embed Model"""
     api_call.post(
@@ -118,10 +128,7 @@ def edit_model(model_type: ModelTypeType, action: Literal["add", "edit"], model_
             key="add_model_name",
             disabled=action == "edit",
         )
-        if model_type == "ll":
-            api_values = list({models["api"] for models in state.ll_model_config.values()})
-        else:
-            api_values = list({models["api"] for models in state.embed_model_config.values()})
+        api_values = get_model_apis(model_type)
         api_index = next((i for i, item in enumerate(api_values) if item == model.api), None)
         model.api = st.selectbox(
             "API:",

src/common/schema.py

@@ -5,7 +5,7 @@
 # spell-checker:ignore ollama, hnsw, mult, ocid, testset
 
 from typing import Optional, Literal, Union
-from pydantic import BaseModel, Field, PrivateAttr
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
 
 from langchain_core.messages import ChatMessage
 import oracledb
@@ -18,6 +18,24 @@
 DistanceMetrics = Literal["COSINE", "EUCLIDEAN_DISTANCE", "DOT_PRODUCT"]
 IndexTypes = Literal["HNSW", "IVF"]
 
+# ModelAPIs
+EmbedAPI = Literal[
+    "OllamaEmbeddings",
+    "OCIGenAIEmbeddings",
+    "CompatOpenAIEmbeddings",
+    "OpenAIEmbeddings",
+    "CohereEmbeddings",
+    "HuggingFaceEndpointEmbeddings",
+]
+LlAPI = Literal[
+    "ChatOllama",
+    "ChatOCIGenAI",
+    "CompatOpenAI",
+    "Perplexity",
+    "OpenAI",
+    "Cohere",
+]
+
 
 #####################################################
 # Database
@@ -110,6 +128,21 @@ class Model(ModelAccess, LanguageModelParameters, EmbeddingModelParameters):
     openai_compat: bool = Field(default=True, description="Is the API OpenAI compatible?")
     status: Statuses = Field(default="UNVERIFIED", description="Status (read-only)", readOnly=True)
 
+    @model_validator(mode="after")
+    def check_api_matches_type(self):
+        """Validate valid API"""
+        ll_apis = LlAPI.__args__
+        embed_apis = EmbedAPI.__args__
+
+        if not self.api or self.api == "unset":
+            return self
+
+        if self.type == "ll" and self.api not in ll_apis:
+            raise ValueError(f"API '{self.api}' is not valid for type 'll'. Must be one of: {ll_apis}")
+        if self.type == "embed" and self.api not in embed_apis:
+            raise ValueError(f"API '{self.api}' is not valid for type 'embed'. Must be one of: {embed_apis}")
+        return self
+
 
 #####################################################
 # Oracle Cloud Infrastructure

src/server/endpoints.py

@@ -12,7 +12,7 @@
 from urllib.parse import urlparse
 from pathlib import Path
 import shutil
-from typing import AsyncGenerator, Literal, Optional
+from typing import AsyncGenerator, Literal, Optional, get_args
 import time
 import requests
 from pydantic import HttpUrl
@@ -309,6 +309,19 @@ async def split_embed(
     #################################################
     # models Endpoints
     #################################################
+    @auth.get("/v1/models/api", description="Get support model APIs", response_model=list)
+    async def models_list_api(
+        model_type: Optional[schema.ModelTypeType] = Query(None),
+    ) -> list[schema.Model]:
+        """List all models APIs after applying filters if specified"""
+        logger.debug("Received models_list_api - type: %s", model_type)
+        if model_type == "ll":
+            return list(get_args(schema.LlAPI))
+        elif model_type == "embed":
+            return list(get_args(schema.EmbedAPI))
+        else:
+            return list()
+
     @auth.get("/v1/models", description="Get all models", response_model=list[schema.Model])
     async def models_list(
         model_type: Optional[schema.ModelTypeType] = Query(None),

tests/server/test_endpoints_models.py

@@ -5,10 +5,11 @@
 # spell-checker: disable
 # pylint: disable=import-error
 
-from typing import Any, Dict
+from typing import Any, Dict, get_args
 import pytest
 from fastapi.testclient import TestClient
 from conftest import TEST_HEADERS, TEST_BAD_HEADERS
+from common.schema import LlAPI, EmbedAPI
 
 
 #############################################################################
@@ -55,6 +56,21 @@ def test_no_auth(self, client: TestClient, test_case: Dict[str, Any]) -> None:
 class TestEndpoints:
     """Test endpoints with AuthN"""
 
+    @pytest.mark.parametrize(
+        "model_type,expected",
+        [
+            ("ll", list(get_args(LlAPI))),
+            ("embed", list(get_args(EmbedAPI))),
+            (None, []),
+        ],
+    )
+    def test_models_list_api(self, client: TestClient, model_type, expected):
+        """Get a list of model APIs to use with tests"""
+        params = {"model_type": model_type} if model_type else {}
+        response = client.get("/v1/models/api", headers=TEST_HEADERS, params=params)
+        assert response.status_code == 200
+        assert sorted(response.json()) == sorted(expected)
+
     def models_list(self, client: TestClient):
         """Get a list of bootstrapped models to use with tests"""
         response = client.get("/v1/models", headers=TEST_HEADERS)