Add llm cache APIs

genai_stack/constant.py

@@ -7,4 +7,5 @@
 VECTORDB = "/vectordb"
 ETL = "/etl"
 PROMPT_ENGINE = "/prompt-engine"
+LLM_CACHE = "/llm-cache"
 MODEL = "/model"

genai_stack/genai_server/models/cache_models.py

@@ -0,0 +1,19 @@
+from pydantic import BaseModel
+
+
+class BaseCacheRequestModel(BaseModel):
+    session_id: int
+    query: str
+    metadata: dict = None
+
+
+class GetCacheRequestModel(BaseCacheRequestModel):
+    pass
+
+
+class SetCacheRequestModel(BaseCacheRequestModel):
+    response: str
+
+
+class CacheResponseModel(BaseCacheRequestModel):
+    response: str

genai_stack/genai_server/routers/cache_routes.py

@@ -0,0 +1,22 @@
+from fastapi import APIRouter
+
+from genai_stack.constant import API, LLM_CACHE
+from genai_stack.genai_server.settings.settings import settings
+from genai_stack.genai_server.models.cache_models import (
+    GetCacheRequestModel, SetCacheRequestModel, CacheResponseModel
+)
+from genai_stack.genai_server.services.cache_service import LLMCacheService
+
+service = LLMCacheService(store=settings.STORE)
+
+router = APIRouter(prefix=API + LLM_CACHE, tags=["llm_cache"])
+
+
+@router.get("/get-cache")
+def get_cache(data: GetCacheRequestModel) -> CacheResponseModel:
+    return service.get_cache(data=data)
+
+
+@router.post("/set-cache")
+def set_cache(data: SetCacheRequestModel) -> CacheResponseModel:
+    return service.set_cache(data=data)

genai_stack/genai_server/server.py

@@ -1,6 +1,7 @@
 from fastapi import FastAPI
 
 from genai_stack.genai_server.routers import (
+    cache_routes,
     session_routes,
     retriever_routes,
     vectordb_routes,
@@ -26,6 +27,7 @@ def get_genai_server_app():
     app.include_router(retriever_routes.router)
     app.include_router(vectordb_routes.router)
     app.include_router(etl_routes.router)
+    app.include_router(cache_routes.router)
     app.include_router(model_routes.router)
 
     return app

genai_stack/genai_server/services/cache_service.py

@@ -0,0 +1,46 @@
+from fastapi import HTTPException
+from sqlalchemy.orm import Session
+
+from genai_stack.genai_platform.services import BaseService
+from genai_stack.genai_server.models.cache_models import GetCacheRequestModel, SetCacheRequestModel, CacheResponseModel
+from genai_stack.genai_server.settings.config import stack_config
+from genai_stack.genai_server.utils import get_current_stack
+from genai_stack.genai_store.schemas import StackSessionSchema
+
+
+class LLMCacheService(BaseService):
+
+    def get_cache(self, data: GetCacheRequestModel) -> CacheResponseModel:
+        with Session(self.engine) as session:
+            stack_session = session.get(StackSessionSchema, data.session_id)
+            if stack_session is None:
+                raise HTTPException(status_code=404, detail=f"Session {data.session_id} not found")
+            stack = get_current_stack(config=stack_config, session=stack_session)
+            response = stack.llm_cache.get_cache(
+                query=data.query,
+                metadata=data.metadata
+            )
+            return CacheResponseModel(
+                session_id=data.session_id,
+                query=data.query,
+                metadata=data.metadata,
+                response=response
+            )
+
+    def set_cache(self, data: SetCacheRequestModel) -> CacheResponseModel:
+        with Session(self.engine) as session:
+            stack_session = session.get(StackSessionSchema, data.session_id)
+            if stack_session is None:
+                raise HTTPException(status_code=404, detail=f"Session {data.session_id} not found")
+            stack = get_current_stack(config=stack_config, session=stack_session)
+            stack.llm_cache.set_cache(
+                query=data.query,
+                response=data.response,
+                metadata=data.metadata
+            )
+            return CacheResponseModel(
+                session_id=data.session_id,
+                query=data.query,
+                metadata=data.metadata,
+                response=data.response
+            )

genai_stack/genai_server/utils/utils.py

@@ -56,7 +56,7 @@ def get_component_class(component_name: str, class_name: str):
 
 # Creating indexes provided by user
 def create_indexes(stack, stack_id: int, session_id: int) -> dict:
-    components = [StackComponentType.VECTOR_DB, StackComponentType.MEMORY]
+    components = [StackComponentType.VECTOR_DB, StackComponentType.MEMORY, StackComponentType.CACHE]
 
     meta_data = {}
     for component in components:
@@ -84,6 +84,7 @@ def get_current_stack(config: dict, session=None, default_session: bool = True):
             if (
                 component_name == StackComponentType.VECTOR_DB.value
                 or component_name == StackComponentType.MEMORY.value
+                or component_name == StackComponentType.CACHE.value
             ):
                 configurations["index_name"] = session.meta_data[component_name]["index_name"]
 

tests/api/test_genai_server/test_cache.py

@@ -0,0 +1,99 @@
+#!/usr/bin/env python
+
+"""Tests for `genai_server`."""
+import unittest
+import requests
+
+
+class TestLLMCacheAPIs(unittest.TestCase):
+
+    def setUp(self) -> None:
+        self.base_url = "http://127.0.0.1:5000/api/llm-cache"
+
+    def test_set_cache(self):
+        response = requests.post(
+            url=self.base_url + "/set-cache",
+            json={
+                "session_id": 1,
+                "query": "Where is sunil from ?",
+                "response": "Sunil is from Hyderabad.",
+                "metadata": {"source": "/path", "page": 1}
+            }
+        )
+        assert response.status_code == 200
+        assert response.json()
+        data = response.json()
+        assert "query" in data.keys()
+        assert "metadata" in data.keys()
+        assert "response" in data.keys()
+
+    def test_get_cache(self):
+        response = requests.get(
+            url=self.base_url + "/get-cache",
+            json={
+                "session_id": 1,
+                "query": "Where is sunil from ?"
+            }
+        )
+
+        assert response.status_code == 200
+        assert response.json()
+        data = response.json()
+        assert "query" in data.keys()
+        assert "metadata" in data.keys()
+        assert "response" in data.keys()
+
+    def test_get_and_set(self):
+        query = "Where is sunil from ?"
+        metadata = {"source": "/path", "page": 1}
+        output = "Sunil is from Hyderabad."
+        response = requests.post(
+            url=self.base_url + "/set-cache",
+            json={
+                "session_id": 1,
+                "query": query,
+                "response": output,
+                "metadata": metadata
+            }
+        )
+        assert response.status_code == 200
+        assert response.json()
+        data = response.json()
+        assert "query" in data.keys() and data.get("query") == query
+        assert "metadata" in data.keys() and data.get("metadata") == metadata
+        assert "response" in data.keys() and data.get("response") == output
+
+        response = requests.get(
+            url=self.base_url + "/get-cache",
+            json={
+                "session_id": 1,
+                "query": query
+            }
+        )
+
+        assert response.status_code == 200
+        assert response.json()
+        data = response.json()
+        assert "query" in data.keys() and data.get("query") == query
+        assert "response" in data.keys() and data.get("response") == output
+
+        response = requests.get(
+            url=self.base_url + "/get-cache",
+            json={
+                "session_id": 1,
+                "query": "Where is sunil from ?",
+                "metadata": {"source": "/pathdiff", "page": 1}
+            }
+        )
+        assert response.status_code != 200
+
+        response = requests.get(
+            url=self.base_url + "/get-cache",
+            json={
+                "session_id": 1,
+                "query": "Where is sunil from ?",
+                "metadata": metadata
+            }
+        )
+
+        assert response.status_code == 200