Changes by aleksandr

src/assistant/lib/agent/openai_functions.py

@@ -36,18 +36,19 @@ class OpenAIFunctions:
                     .order_by(orm_models.FilmWork.embeddings.cosine_distance(embedded_description.root))
                     .limit(5)
                 )
-                neighbours = session.scalars(stmt)
+                response = await session.execute(stmt)
-                for neighbour in await neighbours:
+                neighbours = response.scalars()
+                for neighbour in neighbours:
                     result.append(models.Movie(**neighbour.__dict__))
                 return result
         except sqlalchemy.exc.SQLAlchemyError as error:
             self.logger.exception("Error: %s", error)
 
     @langchain.agents.tool
-    def get_movie_by_id(self, id: uuid.UUID) -> models.Movie | None:
+    def get_movie_by_id(self, id: uuid.UUID = None) -> models.Movie | None:
         """Provide a movie data by movie id."""
-        self.logger.info("Request to get movie by id: %s", id)
+        # self.logger.info("Request to get movie by id: %s", id)
-        return None
+        return f"hello world {id}"
 
     @langchain.agents.tool
     def get_similar_movies(self, id: uuid.UUID) -> list[models.Movie] | None:

src/assistant/lib/agent/repositories.py

@@ -26,7 +26,7 @@ class EmbeddingRepository:
             )  # type: ignore[reportGeneralTypeIssues]
             return models.Embedding(**response["data"][0]["embedding"])
         except openai.error.OpenAIError:
-            self.logger.exception("Failed to get async embedding for: %s", text)
+            self.logger.exception("Failed to get sync embedding for: %s", text)
 
     async def aget_embedding(self, text: str, model: str = "text-embedding-ada-002") -> models.Embedding | None:
         """Get the embedding for a given text.[Async]"""

src/assistant/lib/agent/services.py

@@ -1,5 +1,6 @@
 import asyncio
 import logging
+import typing
 import uuid
 
 import fastapi
@@ -11,7 +12,7 @@ import langchain.prompts
 import langchain.schema
 import langchain.tools.render
 
-import assistant.lib.models.movies as movies
+import lib.models as models
 import lib.agent.openai_functions as openai_functions
 import lib.app.settings as app_settings
 
@@ -23,6 +24,121 @@ class AgentService:
     def __init__(self, settings: app_settings.Settings, tools: openai_functions.OpenAIFunctions) -> None:
         self.settings = settings
         self.tools = tools
+        self.llm = langchain.chat_models.ChatOpenAI(
+            temperature=self.settings.openai.agent_temperature,
+            openai_api_key=self.settings.openai.api_key.get_secret_value()
+        )
+        self.chat_repository = chat_repository
+        self.logger = logging.getLogger(__name__)
+
+    async def get_chat_session_id(self, request: models.RequestLastSessionId) -> uuid.UUID:
+        session_id = self.chat_repository.get_last_session_id(request)
+        if not session_id:
+            session_id = uuid.uuid4()
+        return session_id
+
+    async def artem_process_request(self, request: models.AgentCreateRequestModel) -> models.AgentCreateResponseModel:
+        # Get session ID
+        session_request = models.RequestLastSessionId(
+            channel=request.channel,
+            user_id=request.user_id,
+            minutes_ago=3
+        )
+        session_id = await self.chat_repository.get_last_session_id(session_request)
+        if not session_id:
+            print("NO PREVIOUS CHATS")
+            session_id = uuid.uuid4()
+        print("FOUND CHAT:", )
+        print("SID:", session_id)
+
+        tools = [
+            langchain.tools.Tool(
+                name="GetMovieByDescription",
+                func=self.tools.get_movie_by_description,
+                coroutine=self.tools.get_movie_by_description,
+                description="Get a movie by description"
+            ),
+        ]
+
+        llm = langchain.chat_models.ChatOpenAI(temperature=self.settings.openai.agent_temperature, openai_api_key=self.settings.openai.api_key.get_secret_value())
+
+        chat_history = langchain.memory.ChatMessageHistory()
+        # chat_history = []
+        chat_history_name = f"{chat_history=}".partition("=")[0]
+        request_chat_history = models.RequestChatHistory(session_id=session_id)
+        chat_history_source = await self.chat_repository.get_messages_by_sid(request_chat_history)
+        for entry in chat_history_source:
+            # chat_history.append(langchain.schema.messages.HumanMessage(content=first_question))
+            # chat_history.append(langchain.schema.messages.AIMessage(content=first_result["output"]))
+
+            if entry.content["role"] == "user":
+                chat_history.append(langchain.schema.messages.HumanMessage(content=entry.content["content"]))
+            elif entry.content["role"] == "agent":
+                chat_history.append(langchain.schema.messages.AIMessage(content=entry.content["content"]))
+
+        # chat_history = [entry.model_dump() for entry in chat_history_source]
+        memory_buffer = langchain.memory.ConversationBufferMemory(memory_key=chat_history_name,chat_memory=chat_history)
+
+        print("CHAT HISTORY:", chat_history)
+
+        # chat_history_name = f"{chat_history=}".partition("=")[0]
+
+
+        prompt = langchain.prompts.ChatPromptTemplate.from_messages(
+            [
+                (
+                    "system",
+                    "Act as an advanced AI assistant with extensive capabilities, you have a vast knowledge base about movies and their related aspects. If you are asked about a movie, please use provided functions to retrive data about movies. You can receive a question in any language. Translate it into English. If you don't know the answer, just say that you don't know, don't try to make up an answer. Be concise. ",
+                ),
+                langchain.prompts.MessagesPlaceholder(variable_name=chat_history_name),
+                ("user", "{input}"),
+                langchain.prompts.MessagesPlaceholder(variable_name="agent_scratchpad"),
+            ]
+        )
+
+        llm_with_tools = llm.bind(
+            functions=[langchain.tools.render.format_tool_to_openai_function(tool) for tool in tools]
+        )
+
+        agent = (
+            {
+                "input": lambda _: _["input"],
+                "agent_scratchpad": lambda _: langchain.agents.format_scratchpad.format_to_openai_functions(
+                    _["intermediate_steps"]
+                ),
+                "chat_history": lambda _: _["chat_history"],
+            }
+            | prompt
+            | llm_with_tools
+            | langchain.agents.output_parsers.OpenAIFunctionsAgentOutputParser()
+        )
+
+        agent_executor = langchain.agents.AgentExecutor(agent=agent, tools=tools, verbose=True, memory=memory_buffer)
+        response = await agent_executor.ainvoke({"input": request.text, "chat_history": chat_history})
+        print("AI RESPONSE:", response)
+        user_request = models.RequestChatMessage(
+            session_id=session_id,
+            user_id=request.user_id,
+            channel=request.channel,
+            message={"role": "user", "content": request.text}
+        )
+        ai_response = models.RequestChatMessage(
+            session_id=session_id,
+            user_id=request.user_id,
+            channel=request.channel,
+            message={"role": "assistant", "content": response["output"]}
+        )
+
+        await self.chat_repository.add_message(user_request)
+        await self.chat_repository.add_message(ai_response)
+
+        return response
+
+
+    # TODO: Добавить запрос для процессинга запроса с памятью+
+    # TODO: Улучшить промпт+
+    # TODO: Возможно, надо добавить Chain на перевод
+
 
     async def process_request(self, request: models.AgentCreateRequestModel) -> models.AgentCreateResponseModel:
 
@@ -55,12 +171,56 @@ class AgentService:
         return response_model
 
 
-        return await agent_executor.ainvoke({"input": first_question, "chat_history": chat_history})
+async def main():
+    import lib.agent.repositories as agent_repositories
+    import lib.clients as clients
+
+    postgres_client = clients.AsyncPostgresClient(app_settings.Settings())
+    embedding_repository = agent_repositories.EmbeddingRepository(app_settings.Settings())
+    chat_repository = _chat_repository.ChatHistoryRepository(postgres_client.get_async_session())
+
+    agent_service = AgentService(
+        settings=app_settings.Settings(),
+        tools=openai_functions.OpenAIFunctions(
+            repository=embedding_repository,
+            pg_async_session=postgres_client.get_async_session(),
+        ),
+        chat_repository=chat_repository
+    )
+
+    # question = "What is the movie about a famous country singer meet a talented singer and songwriter who works as a waitress?"
+    request_1  = models.AgentCreateRequestModel(
+        channel="telegram",
+        user_id="123",
+        text="What is the movie about a famous country singer meet a talented singer and songwriter who works as a waitress?"
+    )
+    request_2  = models.AgentCreateRequestModel(
+        channel="telegram",
+        user_id="123",
+        text="So what is the rating of the movie? Do you recommend it?"
+    )
+    request_3  = models.AgentCreateRequestModel(
+        channel="telegram",
+        user_id="123",
+        text="What are the similar movies?"
+    )
+
+    response = await agent_service.artem_process_request(request_1)
+    response = await agent_service.artem_process_request(request_2)
+    response = await agent_service.artem_process_request(request_3)
 
 
-# async def main():
+
+
+    # response = await agent_service.artem_process_request(question)
+    # question = "Highly Rated Titanic Movies"
+    # request = models.AgentCreateRequestModel(text=question)
+    # film_results = await agent_service.process_request(request=request)
+
+    # result = [agent_service.tools.get_movie_by_id(id=film.id) for film in film_results]
+
     # agent_executor = langchain.agents.AgentExecutor(agent=agent, tools=tools, verbose=True)
-
+    #
     # # first_question = "What is the movie where halfling bring the ring to the volcano?"
     # first_question = (
     #     "What is the movie about a famous country singer meet a talented singer and songwriter who works as a waitress?"
@@ -76,5 +236,5 @@ class AgentService:
     # final_result = await agent_executor.ainvoke({"input": third_question, "chat_history": chat_history})
 
 
-# if __name__ == "__main__":
+if __name__ == "__main__":
-#     asyncio.run(main())
+    asyncio.run(main())

src/assistant/lib/api/v1/handlers/voice_responce_handler.py

@@ -5,18 +5,18 @@ import fastapi
 
 import lib.stt.services as stt_services
 
-# import lib.tts.services as tts_service
+import lib.tts.services as tts_service
-# import lib.models as models
+import lib.models as models
 
 
 class VoiceResponseHandler:
     def __init__(
         self,
         stt: stt_services.SpeechService,
-        # tts: tts_service.TTSService,
+        tts: tts_service.TTSService,
     ):
         self.stt = stt
-        # self.tts = tts
+        self.tts = tts
         self.router = fastapi.APIRouter()
         self.router.add_api_route(
             "/",
@@ -36,10 +36,10 @@ class VoiceResponseHandler:
         # TODO: Добавить обработку текста через клиента openai
         # TODO: Добавить синтез речи через клиента tts
         # TODO: Заменить заглушку на реальный ответ
-        # response = await self.tts.get_audio_as_bytes(
+        response = await self.tts.get_audio_as_bytes(
-        #     models.TTSCreateRequestModel(
+            models.TTSCreateRequestModel(
-        #         text=voice_text,
+                text=voice_text,
-        #     )
+            )
-        # )
+        )
-        # return fastapi.responses.StreamingResponse(io.BytesIO(response.audio_content), media_type="audio/ogg")
+        return fastapi.responses.StreamingResponse(io.BytesIO(response.audio_content), media_type="audio/ogg")
-        return fastapi.responses.StreamingResponse(io.BytesIO(voice), media_type="audio/ogg")
+        # return fastapi.responses.StreamingResponse(io.BytesIO(voice), media_type="audio/ogg")

src/assistant/lib/app/app.py

@@ -127,7 +127,7 @@ class Application:
         # TODO: объявить сервисы tts и openai и добавить их в voice_response_handler
         voice_response_handler = api_v1_handlers.VoiceResponseHandler(
             stt=stt_service,
-            # tts=tts_service,  # TODO
+            tts=tts_service,
         ).router
 
         logger.info("Creating application")

src/assistant/lib/app/split_settings/openai.py

@@ -16,3 +16,4 @@ class OpenaiSettings(pydantic_settings.BaseSettings):
         default=..., validation_alias=pydantic.AliasChoices("api_key", "openai_api_key")
     )
     stt_model: str = "whisper-1"
+    agent_temperature: float = 0.7

src/assistant/lib/app/split_settings/postgres.py

@@ -41,4 +41,4 @@ class PostgresSettings(pydantic_settings.BaseSettings):
 
     @property
     def dsn_as_safe_url(self) -> str:
-        return f"{self.driver}://{self.user}:***@{self.host}:{self.port}"
+        return f"{self.driver}://{self.user}:***@{self.host}:{self.port}/{self.db_name}"

src/assistant/lib/models/__init__.py

@@ -3,18 +3,19 @@ from .embedding import Embedding
 from .movies import Movie
 from .token import Token
 from .tts import *
+from .agent import *
 
 
-__all__ = ["Embedding", "Message", "Movie", "RequestChatHistory", "RequestChatMessage", "RequestLastSessionId", "Token"]
+# __all__ = ["Embedding", "Message", "Movie", "RequestChatHistory", "RequestChatMessage", "RequestLastSessionId", "Token"]
 __all__ = [
     "AVAILABLE_MODELS_TYPE",
-    "Base",
+    # "Base",
     "BaseLanguageCodesEnum",
     "BaseVoiceModel",
     "ElevenLabsLanguageCodesEnum",
     "ElevenLabsListVoiceModelsModel",
     "ElevenLabsVoiceModel",
-    "IdCreatedUpdatedBaseMixin",
+    # "IdCreatedUpdatedBaseMixin",
     "LANGUAGE_CODES_ENUM_TYPE",
     "LIST_VOICE_MODELS_TYPE",
     "TTSCreateRequestModel",
@@ -25,4 +26,5 @@ __all__ = [
     "YandexLanguageCodesEnum",
     "YandexListVoiceModelsModel",
     "YandexVoiceModel",
+    "AgentCreateRequestModel",
 ]

src/assistant/pyproject.toml

@@ -13,7 +13,7 @@ profile = "black"
 py_version = "311"
 
 [tool.poetry]
-authors = ["ijaric@gmail.com", "jsdio@jsdio.ru"]
+authors = ["jsdio@jsdio.ru"]
 description = ""
 name = "fastapi_project"
 readme = "README.md"
@@ -22,29 +22,23 @@ version = "0.1.0"
 [tool.poetry.dependencies]
 alembic = "^1.12.0"
 asyncpg = "^0.28.0"
-dill = "^0.3.7"
-faker = "^19.10.0"
 fastapi = "0.103.1"
 greenlet = "^2.0.2"
 httpx = "^0.25.0"
-langchain = "^0.0.312"
-openai = "^0.28.1"
-pgvector = "^0.2.3"
 multidict = "^6.0.4"
-openai = "^0.28.1"
+orjson = "^3.9.7"
-orjson = "3.9.7"
 psycopg2-binary = "^2.9.9"
 pydantic = {extras = ["email"], version = "^2.3.0"}
 pydantic-settings = "^2.0.3"
-pytest = "^7.4.2"
 pytest-asyncio = "^0.21.1"
 python = "^3.11"
 python-jose = "^3.3.0"
-python-magic = "^0.4.27"
-python-multipart = "^0.0.6"
 sqlalchemy = "^2.0.20"
 uvicorn = "^0.23.2"
-wrapt = "^1.15.0"
+pgvector = "^0.2.3"
+python-magic = "^0.4.27"
+openai = "^0.28.1"
+python-multipart = "^0.0.6"
 
 [tool.poetry.dev-dependencies]
 black = "^23.7.0"
@@ -98,9 +92,8 @@ variable-rgx = "^_{0,2}[a-z][a-z0-9_]*$"
 
 [tool.pyright]
 exclude = [
-  ".venv",
-  "alembic"
   ".pytest_cache",
+  ".venv"
 ]
 pythonPlatform = "All"
 pythonVersion = "3.11"