implement get history for image captions, change openai to gpt3.5

app/container/containers.py

@@ -8,7 +8,7 @@
 from botocore.client import BaseClient
 from dependency_injector import containers, providers
 from dependency_injector.providers import Resource
-from langchain.llms import OpenAI
+from langchain.chat_models import ChatOpenAI
 from replicate import Client
 
 from app.infrastructure.auth0.auth0 import Auth0Service
@@ -122,10 +122,10 @@ class Container(containers.DeclarativeContainer):
         model_id=config.infrastructures.replicate.caption_model.model_id,
         caption_client=caption_client,
     )
-    open_ai: OpenAI = providers.Singleton(
-        OpenAI,
+    open_ai: ChatOpenAI = providers.Singleton(
+        ChatOpenAI,
+        model_name=config.infrastructures.open_ai.model_name,
         openai_api_key=config.infrastructures.open_ai.openai_api_key,
-        max_tokens=config.infrastructures.open_ai.max_tokens,
         temperature=config.infrastructures.open_ai.temperature,
     )
 

app/infrastructure/llm/caption.py

@@ -1,7 +1,9 @@
 import logging
 from typing import Generator
 
-from langchain import OpenAI, PromptTemplate
+from langchain import PromptTemplate
+from langchain.chat_models import ChatOpenAI
+from langchain.schema.messages import BaseMessageChunk
 
 
 class ChatGPTCaptionGenerator:
@@ -18,7 +20,7 @@ class ChatGPTCaptionGenerator:
         }
     """
 
-    def __init__(self, model: OpenAI):
+    def __init__(self, model: ChatOpenAI):
         self.logger = logging.getLogger(f"{__name__}.{self.__class__.__name__}")
         self.model = model
 
@@ -35,4 +37,7 @@ def rewrite_caption(
             description=caption,
         )
         for response in self.model.stream(prompt):
-            yield response
+            if isinstance(response, BaseMessageChunk):
+                yield response.dict()["content"]
+            elif isinstance(response, str):
+                yield response

app/infrastructure/llm/vocabulary.py

@@ -3,7 +3,9 @@
 from datetime import datetime
 from typing import Generator
 
-from langchain import OpenAI, PromptTemplate
+from langchain import PromptTemplate
+from langchain.chat_models import ChatOpenAI
+from langchain.schema.messages import BaseMessageChunk
 
 
 class ChatGPTVocabularyGenerator:
@@ -32,7 +34,7 @@ class ChatGPTVocabularyGenerator:
         '{"{{ learning_language }}": {{ format_output }}, "{{ primary_language }}": {{ format_output }}}'
     )
 
-    def __init__(self, model: OpenAI):
+    def __init__(self, model: ChatOpenAI):
         self.logger = logging.getLogger(f"{__name__}.{self.__class__.__name__}")
         self.model = model
 
@@ -63,8 +65,11 @@ def generate_vocabulary_questions(
         prompt = textwrap.dedent(prompt)
         self.logger.debug(f"Request: {prompt}")
         for text in self.model.stream(prompt):
+            if isinstance(text, BaseMessageChunk):
+                yield text.dict()["content"]
+            elif isinstance(text, str):
+                yield text
             response += text
-            yield text
 
         self.logger.debug(f"Execution time: {datetime.now() - start}")
         self.logger.debug(f"Response: {response}")

app/repositories/caption_repository.py

@@ -1,7 +1,7 @@
 from contextlib import AbstractContextManager
-from typing import Callable
+from typing import Callable, List
 
-from sqlalchemy.orm import Session
+from sqlalchemy.orm import Session, joinedload
 
 from app.models.db.image_caption import (
     ImageCaptionLearningLanguage,
@@ -35,6 +35,24 @@ def add_image_caption(
             session.commit()
             return img_caption
 
+    def get_caption(self, user_id: str) -> List[ImageCaptionPrimaryLanguage]:
+        with self.session_factory() as session:
+            query = (
+                session.query(ImageCaptionPrimaryLanguage)
+                .join(ImageCaptionLearningLanguage)
+                .options(
+                    joinedload(ImageCaptionPrimaryLanguage.primary_language),
+                    joinedload(ImageCaptionPrimaryLanguage.learning_caption).joinedload(
+                        ImageCaptionLearningLanguage.learning_language
+                    ),
+                )
+                .filter(
+                    ImageCaptionPrimaryLanguage.user_id == user_id,
+                )
+                .order_by(ImageCaptionPrimaryLanguage.time_created.desc())
+            )
+            return query.all()
+
 
 class TranslatedCaptionRepository:
     def __init__(
@@ -44,10 +62,13 @@ def __init__(
         self.session_factory = session_factory
 
     def add_translated_caption(
-        self, learning_caption: str, image_caption_object: ImageCaptionPrimaryLanguage
+        self,
+        learning_caption: str,
+        learning_language: str,
+        image_caption_object: ImageCaptionPrimaryLanguage,
     ) -> ImageCaptionLearningLanguage:
         with self.session_factory() as session:
-            learning_language_id = check_language(session, learning_caption)
+            learning_language_id = check_language(session, learning_language)
             translated_image_caption = ImageCaptionLearningLanguage(
                 learning_language_id=learning_language_id,
                 learning_language_caption=learning_caption,

app/routes/api_v1/endpoints/image.py

@@ -105,3 +105,12 @@ async def generate_caption(
         ),
         media_type="text/plain",
     )
+
+
+@router.get("/caption")
+@inject
+async def get_caption_history(
+    auth: Auth0User = Depends(check_user),
+    caption_service: CaptionService = Depends(Provide[Container.caption_service]),
+) -> object:
+    return caption_service.list_caption_history(user_id=auth.id)

app/services/caption_service.py

@@ -1,8 +1,9 @@
 import json
-from typing import Generator
+from typing import Generator, List
 
 from app.infrastructure.llm.caption import ChatGPTCaptionGenerator
 from app.infrastructure.replicate.caption import CaptionGenerator
+from app.models.db.image_caption import ImageCaptionPrimaryLanguage
 from app.models.schemas.image_caption import ImageCaptionCreate
 from app.repositories.caption_repository import (
     CaptionRepository,
@@ -46,5 +47,9 @@ def get_caption_from_image(self, user_id: str, caption_input: dict) -> Generator
         )
         self.learning_caption_repository.add_translated_caption(
             learning_caption=caption_data[caption_input["learning_language"]],
+            learning_language=caption_input["learning_language"],
             image_caption_object=caption_insert_object,
         )
+
+    def list_caption_history(self, user_id: str) -> List[ImageCaptionPrimaryLanguage]:
+        return self.primary_caption_repository.get_caption(user_id=user_id)

app/tests/image_caption/test_image_caption.py

@@ -5,7 +5,7 @@
 
 import pytest
 from fastapi.testclient import TestClient
-from langchain import OpenAI
+from langchain.chat_models import ChatOpenAI
 from replicate import Client
 
 from app.infrastructure.auth0.auth0 import Auth0Service
@@ -65,7 +65,7 @@ def test_generate_caption(client):
 
     primary_caption_repository_mock = mock.AsyncMock(spec=CaptionRepository)
     learning_caption_repository_mock = mock.AsyncMock(spec=TranslatedCaptionRepository)
-    open_ai_mock = mock.Mock(spec=OpenAI)
+    open_ai_mock = mock.Mock(spec=ChatOpenAI)
     open_ai_mock.stream.return_value = mock_chat_gpt_response()
 
     replicate_caption_mock = mock.Mock(spec=Client)
@@ -110,7 +110,7 @@ def test_generate_caption_fail(client):
 
     primary_caption_repository_mock = mock.AsyncMock(spec=CaptionRepository)
     learning_caption_repository_mock = mock.AsyncMock(spec=TranslatedCaptionRepository)
-    open_ai_mock = mock.Mock(spec=OpenAI)
+    open_ai_mock = mock.Mock(spec=ChatOpenAI)
     open_ai_mock.stream.return_value = mock_fail_response()
 
     replicate_caption_mock = mock.Mock(spec=Client)

app/tests/vocabulary/test_vocabulary.py

@@ -4,7 +4,7 @@
 
 import pytest
 from fastapi.testclient import TestClient
-from langchain import OpenAI
+from langchain.chat_models import ChatOpenAI
 
 from app.main import app
 from app.models.db.vocabulary import Category
@@ -149,7 +149,7 @@ def test_func_generate_vocabulary_questions(client):
     auth_service_mock = mock_user()
     voca_repo_mock = mock.Mock(spec=VocabularyRepository)
 
-    open_ai_mock = mock.Mock(spec=OpenAI)
+    open_ai_mock = mock.Mock(spec=ChatOpenAI)
     open_ai_mock.stream.return_value = mock_chat_gpt_response()
 
     app.container.auth.override(auth_service_mock)
@@ -190,7 +190,7 @@ def test_func_generate_vocabulary_questions_invalid_level_type(client):
 def test_prompt_parse_failed(client):
     auth_service_mock = mock_user()
     voca_repo_mock = mock.Mock(spec=VocabularyRepository)
-    open_ai_mock = mock.Mock(spec=OpenAI)
+    open_ai_mock = mock.Mock(spec=ChatOpenAI)
     open_ai_mock.stream.return_value = mock_wrong_chat_gpt_response()
     app.container.auth.override(auth_service_mock)
     app.container.open_ai.override(open_ai_mock)

config.yml

@@ -9,8 +9,8 @@ core:
 
 infrastructures:
   open_ai:
+    model_name: ${OPENAI_MODEL_NAME:"gpt-3.5-turbo"}
     openai_api_key: ${OPENAI_API_KEY}
-    max_tokens: ${OPENAI_MAX_TOKENS:"-1"}
     temperature: ${OPENAI_TEMPERATURE:"0.5"}
   aws:
     access_key_id: ${AWS_ACCESS_KEY_ID}