devops/encoach_backend
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merged in release/async (pull request #40)

Browse Source 
Release/async

Approved-by: Tiago Ribeiro
This commit is contained in:
carlos.mesquita
2024-11-26 10:28:02 +00:00
committed by Tiago Ribeiro
parent c74b2b9b7b 72d2c0121a
commit 54a01f9631
21 changed files with 525 additions and 199 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
app/api/grade.py
View File

@@ -1,9 +1,8 @@
from dependency_injector.wiring import inject, Provide
from fastapi import APIRouter, Depends, Path, Request
from fastapi import APIRouter, Depends, Path, Request, BackgroundTasks
from app.controllers.abc import IGradeController
from app.dtos.writing import WritingGradeTaskDTO
from app.dtos.speaking import GradeSpeakingAnswersDTO, GradeSpeakingDTO
from app.middlewares import Authorized, IsAuthenticatedViaBearerToken
controller = "grade_controller"
@@ -17,35 +16,51 @@ grade_router = APIRouter()
@inject
async def grade_writing_task(
data: WritingGradeTaskDTO,
background_tasks: BackgroundTasks,
task: int = Path(..., ge=1, le=2),
grade_controller: IGradeController = Depends(Provide[controller])
):
return await grade_controller.grade_writing_task(task, data)
return await grade_controller.grade_writing_task(task, data, background_tasks)
@grade_router.post(
'/speaking/2',
dependencies=[Depends(Authorized([IsAuthenticatedViaBearerToken]))]
)
@inject
async def grade_speaking_task_2(
data: GradeSpeakingDTO,
grade_controller: IGradeController = Depends(Provide[controller])
):
return await grade_controller.grade_speaking_task(2, [data.dict()])
@grade_router.post(
'/speaking/{task}',
dependencies=[Depends(Authorized([IsAuthenticatedViaBearerToken]))]
)
@inject
async def grade_speaking_task_1_and_3(
data: GradeSpeakingAnswersDTO,
async def grade_speaking_task(
request: Request,
background_tasks: BackgroundTasks,
task: int = Path(..., ge=1, le=3),
grade_controller: IGradeController = Depends(Provide[controller])
):
return await grade_controller.grade_speaking_task(task, data.answers)
form = await request.form()
return await grade_controller.grade_speaking_task(task, form, background_tasks)
@grade_router.get(
'/pending/{sessionId}',
dependencies=[Depends(Authorized([IsAuthenticatedViaBearerToken]))]
)
@inject
async def get_pending_evaluations(
session_id: str,
grade_controller: IGradeController = Depends(Provide[controller])
):
return await grade_controller.get_evaluations(session_id, "pending")
@grade_router.get(
'/completed/{sessionId}',
dependencies=[Depends(Authorized([IsAuthenticatedViaBearerToken]))]
)
@inject
async def get_completed_evaluations(
session_id: str,
grade_controller: IGradeController = Depends(Provide[controller])
):
return await grade_controller.get_evaluations(session_id, "completed")
@grade_router.post(

1
app/api/level.py
View File

@@ -18,7 +18,6 @@ async def generate_exercises(
dto: LevelExercisesDTO,
level_controller: ILevelController = Depends(Provide[controller])
):
print(dto.dict())
return await level_controller.generate_exercises(dto)
@level_router.get(

8
app/api/speaking.py
View File

@@ -3,9 +3,8 @@ from typing import Optional
from dependency_injector.wiring import inject, Provide
from fastapi import APIRouter, Path, Query, Depends
from pydantic import BaseModel
from app.dtos.video import Task, TaskStatus
from app.dtos.speaking import Video
from app.middlewares import Authorized, IsAuthenticatedViaBearerToken
from app.configs.constants import EducationalContent
from app.controllers.abc import ISpeakingController
@@ -13,11 +12,6 @@ from app.controllers.abc import ISpeakingController
controller = "speaking_controller"
speaking_router = APIRouter()
class Video(BaseModel):
text: str
avatar: str
@speaking_router.post(
'/media',
dependencies=[Depends(Authorized([IsAuthenticatedViaBearerToken]))]

27
app/configs/dependency_injection.py
View File

@@ -10,19 +10,21 @@ from dotenv import load_dotenv
from sentence_transformers import SentenceTransformer
from app.repositories.impl import *
from app.repositories.impl.document_stores.mongo import MongoDB
from app.services.impl import *
from app.controllers.impl import *
from app.services.impl.exam.evaluation import EvaluationService
load_dotenv()
class DependencyInjector:
def __init__(self, polly_client: any, http_client: HTTPClient, whisper_model: any):
def __init__(self, polly_client: any, http_client: HTTPClient, stt: OpenAIWhisper):
self._container = containers.DynamicContainer()
self._polly_client = polly_client
self._http_client = http_client
self._whisper_model = whisper_model
self._stt = stt
def inject(self):
self._setup_clients()
@@ -33,22 +35,25 @@ class DependencyInjector:
self._container.wire(
packages=["app"]
)
return self
def _setup_clients(self):
self._container.openai_client = providers.Singleton(AsyncOpenAI)
self._container.polly_client = providers.Object(self._polly_client)
self._container.http_client = providers.Object(self._http_client)
self._container.whisper_model = providers.Object(self._whisper_model)
self._container.stt = providers.Object(self._stt)
def _setup_third_parties(self):
self._container.llm = providers.Factory(OpenAI, client=self._container.openai_client)
self._container.stt = providers.Factory(OpenAIWhisper, model=self._container.whisper_model)
self._container.tts = providers.Factory(AWSPolly, client=self._container.polly_client)
"""
with open('app/services/impl/third_parties/elai/conf.json', 'r') as file:
elai_conf = json.load(file)
with open('app/services/impl/third_parties/elai/avatars.json', 'r') as file:
elai_avatars = json.load(file)
"""
with open('app/services/impl/third_parties/heygen/avatars.json', 'r') as file:
heygen_avatars = json.load(file)
@@ -64,8 +69,8 @@ class DependencyInjector:
cred = credentials.Certificate(os.getenv("GOOGLE_APPLICATION_CREDENTIALS"))
firebase_token = cred.get_access_token().access_token
self._container.document_store = providers.Object(
AsyncIOMotorClient(os.getenv("MONGODB_URI"))[os.getenv("MONGODB_DB")]
self._container.document_store = providers.Factory(
MongoDB, mongo_db=AsyncIOMotorClient(os.getenv("MONGODB_URI"))[os.getenv("MONGODB_DB")]
)
self._container.firebase_instance = providers.Factory(
@@ -123,11 +128,17 @@ class DependencyInjector:
UserService, document_store=self._container.document_store
)
self._container.evaluation_service = providers.Factory(
EvaluationService, db=self._container.document_store,
writing_service=self._container.writing_service,
speaking_service=self._container.speaking_service
)
def _setup_controllers(self):
self._container.grade_controller = providers.Factory(
GradeController, grade_service=self._container.grade_service,
speaking_service=self._container.speaking_service,
writing_service=self._container.writing_service
evaluation_service=self._container.evaluation_service
)
self._container.user_controller = providers.Factory(

18
app/controllers/abc/grade.py
View File

@@ -1,15 +1,27 @@
from abc import ABC, abstractmethod
from typing import Dict, List
from typing import Dict, List, Union
from fastapi import BackgroundTasks
from fastapi.datastructures import FormData
class IGradeController(ABC):
@abstractmethod
async def grade_writing_task(self, task: int, data):
async def grade_writing_task(
self,
task: int, dto: any,
background_tasks: BackgroundTasks
):
pass
@abstractmethod
async def grade_speaking_task(self, task: int, answers: List[Dict]) -> Dict:
async def grade_speaking_task(
self, task: int, form: FormData, background_tasks: BackgroundTasks
):
pass
@abstractmethod
async def get_evaluations(self, session_id: str, status: str):
pass
@abstractmethod

95
app/controllers/impl/grade.py
View File

@@ -1,34 +1,97 @@
import logging
from typing import Dict, List
from typing import Dict, List, Union
from uuid import uuid4
from fastapi import BackgroundTasks, Response, HTTPException
from fastapi.datastructures import FormData
from app.configs.constants import FilePaths
from app.controllers.abc import IGradeController
from app.dtos.evaluation import EvaluationType
from app.dtos.speaking import GradeSpeakingItem
from app.dtos.writing import WritingGradeTaskDTO
from app.helpers import FileHelper
from app.services.abc import ISpeakingService, IWritingService, IGradeService
from app.utils import handle_exception
from app.services.abc import IGradeService, IEvaluationService
class GradeController(IGradeController):
def __init__(
self,
grade_service: IGradeService,
speaking_service: ISpeakingService,
writing_service: IWritingService
evaluation_service: IEvaluationService,
):
self._service = grade_service
self._speaking_service = speaking_service
self._writing_service = writing_service
self._evaluation_service = evaluation_service
self._logger = logging.getLogger(__name__)
async def grade_writing_task(self, task: int, data: WritingGradeTaskDTO):
return await self._writing_service.grade_writing_task(task, data.question, data.answer)
async def grade_writing_task(
self,
task: int, dto: WritingGradeTaskDTO, background_tasks: BackgroundTasks
):
await self._evaluation_service.create_evaluation(
dto.userId, dto.sessionId, dto.exerciseId, EvaluationType.WRITING, task
)
@handle_exception(400)
async def grade_speaking_task(self, task: int, answers: List[Dict]) -> Dict:
FileHelper.delete_files_older_than_one_day(FilePaths.AUDIO_FILES_PATH)
return await self._speaking_service.grade_speaking_task(task, answers)
await self._evaluation_service.begin_evaluation(
dto.userId, dto.sessionId, task, dto.exerciseId, EvaluationType.WRITING, dto, background_tasks
)
return Response(status_code=200)
async def grade_speaking_task(self, task: int, form: FormData, background_tasks: BackgroundTasks):
answers: Dict[int, Dict] = {}
user_id = form.get("userId")
session_id = form.get("sessionId")
exercise_id = form.get("exerciseId")
if not session_id or not exercise_id:
raise HTTPException(
status_code=400,
detail="Fields sessionId and exerciseId are required!"
)
for key, value in form.items():
if '_' not in key:
continue
field_name, index = key.rsplit('_', 1)
index = int(index)
if index not in answers:
answers[index] = {}
if field_name == 'question':
answers[index]['question'] = value
elif field_name == 'audio':
answers[index]['answer'] = value
for i, answer in answers.items():
if 'question' not in answer or 'answer' not in answer:
raise HTTPException(
status_code=400,
detail=f"Incomplete data for answer {i}. Both question and audio required."
)
items = [
GradeSpeakingItem(
question=answers[i]['question'],
answer=answers[i]['answer']
)
for i in sorted(answers.keys())
]
ex_type = EvaluationType.SPEAKING if task == 2 else EvaluationType.SPEAKING_INTERACTIVE
await self._evaluation_service.create_evaluation(
user_id, session_id, exercise_id, ex_type, task
)
await self._evaluation_service.begin_evaluation(
user_id, session_id, task, exercise_id, ex_type, items, background_tasks
)
return Response(status_code=200)
async def get_evaluations(self, session_id: str, status: str):
return await self._evaluation_service.get_evaluations(session_id, status)
async def grade_short_answers(self, data: Dict):
return await self._service.grade_short_answers(data)

18
app/dtos/evaluation.py Normal file
View File

@@ -0,0 +1,18 @@
from enum import Enum
from typing import Dict, Optional
from pydantic import BaseModel
class EvaluationType(str, Enum):
WRITING = "writing"
SPEAKING_INTERACTIVE = "speaking_interactive"
SPEAKING = "speaking"
class EvaluationRecord(BaseModel):
id: str
session_id: str
exercise_id: str
type: EvaluationType
task: int
status: str = "pending"
result: Optional[Dict] = None

28
app/dtos/speaking.py
View File

@@ -1,27 +1,13 @@
import random
from typing import List, Dict, Optional
from typing import List, Dict
from fastapi import UploadFile
from pydantic import BaseModel
from app.configs.constants import MinTimers
class Video(BaseModel):
text: str
avatar: str
class SaveSpeakingDTO(BaseModel):
exercises: List[Dict]
minTimer: int = MinTimers.SPEAKING_MIN_TIMER_DEFAULT
class GradeSpeakingDTO(BaseModel):
class GradeSpeakingItem(BaseModel):
question: str
answer: str
class GradeSpeakingAnswersDTO(BaseModel):
answers: List[Dict]
class GenerateVideo1DTO(BaseModel):
avatar: str = Optional[str]
questions: List[str]
first_topic: str
second_topic: str
answer: UploadFile

3
app/dtos/writing.py
View File

@@ -2,5 +2,8 @@ from pydantic import BaseModel
class WritingGradeTaskDTO(BaseModel):
userId: str
sessionId: str
exerciseId: str
question: str
answer: str

4
app/repositories/impl/document_stores/mongo.py
View File

@@ -34,8 +34,8 @@ class MongoDB(IDocumentStore):
cursor = self._mongo_db[collection].find(query)
return [document async for document in cursor]
async def update(self, collection: str, filter_query: Dict, update: Dict) -> Optional[str]:
return (await self._mongo_db[collection].update_one(filter_query, update)).upserted_id
async def update(self, collection: str, filter_query: Dict, update: Dict) -> Optional[str]:
return (await self._mongo_db[collection].update_one(filter_query, update, upsert=True)).upserted_id
async def get_doc_by_id(self, collection: str, doc_id: str) -> Optional[Dict]:
return await self._mongo_db[collection].find_one({"id": doc_id})

3
app/repositories/impl/file_storage/firebase.py
View File

@@ -55,8 +55,7 @@ class FirebaseStorage(IFileStorage):
if response.status_code == 200:
self._logger.info(f"File {source_file_name} uploaded to {self._storage_url}/o/{destination_blob_name}.")
# TODO: Test this
#await self.make_public(destination_blob_name)
await self.make_public(destination_blob_name)
file_url = f"{self._storage_url}/o/{destination_blob_name}"
return file_url

8
app/server.py
View File

@@ -12,7 +12,6 @@ from typing import List
from http import HTTPStatus
import httpx
import whisper
from fastapi import FastAPI, Request
from fastapi.encoders import jsonable_encoder
from fastapi.exceptions import RequestValidationError
@@ -27,6 +26,7 @@ from app.api import router
from app.configs import DependencyInjector
from app.exceptions import CustomException
from app.middlewares import AuthenticationMiddleware, AuthBackend
from app.services.impl import OpenAIWhisper
@asynccontextmanager
@@ -36,8 +36,6 @@ async def lifespan(_app: FastAPI):
https://fastapi.tiangolo.com/advanced/events/
"""
# Whisper model
whisper_model = whisper.load_model("base")
# NLTK required datasets download
nltk.download('words')
@@ -56,11 +54,12 @@ async def lifespan(_app: FastAPI):
)
http_client = httpx.AsyncClient()
stt = OpenAIWhisper()
DependencyInjector(
polly_client,
http_client,
whisper_model
stt
).inject()
# Setup logging
@@ -72,6 +71,7 @@ async def lifespan(_app: FastAPI):
yield
stt.close()
await http_client.aclose()
await polly_client.close()
await context_stack.aclose()

4
app/services/abc/__init__.py
View File

@@ -2,9 +2,11 @@ from .third_parties import *
from .exam import *
from .training import *
from .user import IUserService
from .evaluation import IEvaluationService
__all__ = [
"IUserService"
"IUserService",
"IEvaluationService"
]
__all__.extend(third_parties.__all__)
__all__.extend(exam.__all__)

34
app/services/abc/evaluation.py Normal file
View File

@@ -0,0 +1,34 @@
from abc import abstractmethod, ABC
from typing import Union, List, Dict
from fastapi import BackgroundTasks
from app.dtos.evaluation import EvaluationType
class IEvaluationService(ABC):
@abstractmethod
async def create_evaluation(
self,
user_id: str,
session_id: str,
exercise_id: str,
eval_type: EvaluationType,
task: int
):
pass
@abstractmethod
async def begin_evaluation(
self,
user_id: str, session_id: str, task: int,
exercise_id: str, exercise_type: str,
solution: any,
background_tasks: BackgroundTasks
):
pass
@abstractmethod
async def get_evaluations(self, session_id: str, status: str) -> List[Dict]:
pass

2
app/services/abc/exam/speaking.py
View File

@@ -11,6 +11,6 @@ class ISpeakingService(ABC):
pass
@abstractmethod
async def grade_speaking_task(self, task: int, answers: List[Dict]) -> Dict:
async def grade_speaking_task(self, task: int, items: any) -> Dict:
pass

112
app/services/impl/exam/evaluation.py Normal file
View File

@@ -0,0 +1,112 @@
import logging
from typing import Union, Dict, List
from fastapi import BackgroundTasks
from app.dtos.evaluation import EvaluationType
from app.dtos.speaking import GradeSpeakingItem
from app.dtos.writing import WritingGradeTaskDTO
from app.repositories.abc import IDocumentStore
from app.services.abc import IWritingService, ISpeakingService, IEvaluationService
class EvaluationService(IEvaluationService):
def __init__(self, db: IDocumentStore, writing_service: IWritingService, speaking_service: ISpeakingService):
self._db = db
self._writing_service = writing_service
self._speaking_service = speaking_service
self._logger = logging.getLogger(__name__)
async def create_evaluation(
self,
user_id: str,
session_id: str,
exercise_id: str,
eval_type: EvaluationType,
task: int
):
await self._db.save_to_db(
"evaluation",
{
"user": user_id,
"session_id": session_id,
"exercise_id": exercise_id,
"type": eval_type,
"task": task,
"status": "pending"
}
)
async def begin_evaluation(
self,
user_id: str, session_id: str, task: int,
exercise_id: str, exercise_type: str,
solution: Union[WritingGradeTaskDTO, List[GradeSpeakingItem]],
background_tasks: BackgroundTasks
):
background_tasks.add_task(
self._begin_evaluation,
user_id, session_id, task,
exercise_id, exercise_type,
solution
)
async def _begin_evaluation(
self, user_id: str, session_id: str, task: int,
exercise_id: str, exercise_type: str,
solution: Union[WritingGradeTaskDTO, List[GradeSpeakingItem]]
):
try:
if exercise_type == EvaluationType.WRITING:
result = await self._writing_service.grade_writing_task(
task,
solution.question,
solution.answer
)
else:
result = await self._speaking_service.grade_speaking_task(
task,
solution
)
await self._db.update(
"evaluation",
{
"user": user_id,
"exercise_id": exercise_id,
"session_id": session_id,
},
{
"$set": {
"status": "completed",
"result": result,
}
}
)
except Exception as e:
self._logger.error(f"Error processing evaluation {session_id} - {exercise_id}: {str(e)}")
await self._db.update(
"evaluation",
{
"user": user_id,
"exercise_id": exercise_id,
"session_id": session_id
},
{
"$set": {
"status": "error",
"error": str(e),
}
}
)
async def get_evaluations(self, session_id: str, status: str) -> List[Dict]:
return await self._db.find(
"evaluation",
{
"session_id": session_id,
"status": status
}
)

28
app/services/impl/exam/level/upload.py
View File

@@ -1,8 +1,10 @@
from uuid import uuid4
import aiofiles
import os
from logging import getLogger
from typing import Dict, Any, Coroutine, Optional
from typing import Dict, Any, Optional
import pdfplumber
from fastapi import UploadFile
@@ -21,20 +23,19 @@ class UploadLevelModule:
self._logger = getLogger(__name__)
self._llm = openai
async def generate_level_from_file(self, file: UploadFile, solutions: Optional[UploadFile]) -> Dict[str, Any] | None:
ext, path_id = await FileHelper.save_upload(file)
FileHelper.convert_file_to_pdf(
f'./tmp/{path_id}/upload.{ext}', f'./tmp/{path_id}/exercises.pdf'
)
file_has_images = self._check_pdf_for_images(f'./tmp/{path_id}/exercises.pdf')
async def generate_level_from_file(self, exercises: UploadFile, solutions: Optional[UploadFile]) -> Dict[str, Any] | None:
path_id = str(uuid4())
ext, _ = await FileHelper.save_upload(exercises, "exercises", path_id)
FileHelper.convert_file_to_html(f'./tmp/{path_id}/exercises.{ext}', f'./tmp/{path_id}/exercises.html')
if not file_has_images:
FileHelper.convert_file_to_html(f'./tmp/{path_id}/upload.{ext}', f'./tmp/{path_id}/exercises.html')
if solutions:
ext, _ = await FileHelper.save_upload(solutions, "solutions", path_id)
FileHelper.convert_file_to_html(f'./tmp/{path_id}/solutions.{ext}', f'./tmp/{path_id}/solutions.html')
completion: Coroutine[Any, Any, Exam] = (
self._png_completion(path_id) if file_has_images else self._html_completion(path_id)
)
response = await completion
#completion: Coroutine[Any, Any, Exam] = (
# self._png_completion(path_id) if file_has_images else self._html_completion(path_id)
#)
response = await self._html_completion(path_id)
FileHelper.remove_directory(f'./tmp/{path_id}')
@@ -42,6 +43,7 @@ class UploadLevelModule:
return self.fix_ids(response.model_dump(exclude_none=True))
return None
@staticmethod
@suppress_loggers()
def _check_pdf_for_images(pdf_path: str) -> bool:

2
app/services/impl/exam/reading/import_reading.py
View File

@@ -98,7 +98,7 @@ class ImportReadingModule:
]
}
],
"text": "<numbered questions with format: <question text>{{<question number>}}\\n>",
"text": "<numbered questions with format in square brackets: [<question text>{{<question number>}}\\\\n] notice how there is a double backslash before the n -> I want an escaped newline in your output> ",
"type": "writeBlanks",
"prompt": "<specific instructions for this exercise section>"
}

177
app/services/impl/exam/speaking.py
View File

@@ -1,6 +1,9 @@
import asyncio
import logging
import os
import random
from uuid import uuid4
import aiofiles
import re
import uuid
from typing import Dict, List, Optional
@@ -9,6 +12,7 @@ from app.configs.constants import (
FieldsAndExercises, GPTModels, TemperatureSettings,
FilePaths
)
from app.dtos.speaking import GradeSpeakingItem
from app.helpers import TextHelper
from app.repositories.abc import IFileStorage, IDocumentStore
from app.services.abc import ISpeakingService, ILLMService, IVideoGeneratorService, ISpeechToTextService
@@ -165,105 +169,120 @@ class SpeakingService(ISpeakingService):
return response
async def grade_speaking_task(self, task: int, answers: List[Dict]) -> Dict:
request_id = uuid.uuid4()
self._logger.info(
f'POST - speaking_task_{task} - Received request to grade speaking task {task}. '
f'Use this id to track the logs: {str(request_id)} - Request data: {str(answers)}'
)
text_answers = []
perfect_answers = []
async def grade_speaking_task(self, task: int, items: List[GradeSpeakingItem]) -> Dict:
request_id = str(uuid.uuid4())
self._log(task, request_id, f"Received request to grade speaking task {task}.")
if task != 2:
self._logger.info(
f'POST - speaking_task_{task} - {str(request_id)} - Received {str(len(answers))} total answers.'
)
self._log(task, request_id, f'Received {len(items)} total answers.')
for item in answers:
sound_file_name = FilePaths.AUDIO_FILES_PATH + str(uuid.uuid4())
temp_files = []
try:
# Save all files first
temp_files = await asyncio.gather(*[
self.save_file(item) for item in items
])
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Downloading file {item["answer"]}')
# Process all transcriptions concurrently (up to 4)
self._log(task, request_id, 'Starting batch transcription')
text_answers = await asyncio.gather(*[
self._stt.speech_to_text(file_path)
for file_path in temp_files
])
await self._file_storage.download_firebase_file(item["answer"], sound_file_name)
for answer in text_answers:
self._log(task, request_id, f'Transcribed answer: {answer}')
if not TextHelper.has_x_words(answer, 20):
self._log(
task, request_id,
f'The answer had less words than threshold 20 to be graded. Answer: {answer}'
)
return self._zero_rating("The audio recorded does not contain enough english words to be graded.")
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - '
f'Downloaded file {item["answer"]} to {sound_file_name}'
)
# Get perfect answers
self._log(task, request_id, 'Requesting perfect answers')
perfect_answers = await asyncio.gather(*[
self._get_perfect_answer(task, item.question)
for item in items
])
answer_text = await self._stt.speech_to_text(sound_file_name)
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Transcripted answer: {answer_text}')
# Format the responses
if task in {1, 3}:
self._log(task, request_id, 'Formatting answers and questions for prompt.')
text_answers.append(answer_text)
item["answer"] = answer_text
os.remove(sound_file_name)
formatted_text = ""
for i, (item, transcribed_answer) in enumerate(zip(items, text_answers), start=1):
formatted_text += f"**Question {i}:**\n{item.question}\n\n"
formatted_text += f"**Answer {i}:**\n{transcribed_answer}\n\n"
# TODO: This will end the grading of all answers if a single one does not have enough words
# don't know if this is intended
if not TextHelper.has_x_words(answer_text, 20):
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - '
f'The answer had less words than threshold 20 to be graded. Answer: {answer_text}'
)
return self._zero_rating("The audio recorded does not contain enough english words to be graded.")
self._log(task, request_id, f'Formatted answers and questions for prompt: {formatted_text}')
questions_and_answers = f'\n\n The questions and answers are: \n\n{formatted_text}'
else:
questions_and_answers = f'\n Question: "{items[0].question}" \n Answer: "{text_answers[0]}"'
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - '
f'Requesting perfect answer for question: {item["question"]}'
)
perfect_answers.append(await self._get_perfect_answer(task, item["question"]))
self._log(task, request_id, 'Requesting grading of the answer(s).')
response = await self._grade_task(task, questions_and_answers)
self._log(task, request_id, f'Answer(s) graded: {response}')
if task in {1, 3}:
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - Formatting answers and questions for prompt.'
)
if task in {1, 3}:
self._log(task, request_id, 'Adding perfect answer(s) to response.')
formatted_text = ""
for i, entry in enumerate(answers, start=1):
formatted_text += f"**Question {i}:**\n{entry['question']}\n\n"
formatted_text += f"**Answer {i}:**\n{entry['answer']}\n\n"
# TODO: check if it is answer["answer"] instead
for i, answer in enumerate(perfect_answers, start=1):
response['perfect_answer_' + str(i)] = answer
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - '
f'Formatted answers and questions for prompt: {formatted_text}'
)
questions_and_answers = f'\n\n The questions and answers are: \n\n{formatted_text}'
else:
questions_and_answers = f'\n Question: "{answers[0]["question"]}" \n Answer: "{answers[0]["answer"]}"'
self._log(task, request_id, 'Getting speaking corrections in parallel')
# Get all corrections in parallel
fixed_texts = await asyncio.gather(*[
self._get_speaking_corrections(answer)
for answer in text_answers
])
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Requesting grading of the answer(s).')
response = await self._grade_task(task, questions_and_answers)
self._log(task, request_id, 'Adding transcript and fixed texts to response.')
for i, (answer, fixed) in enumerate(zip(text_answers, fixed_texts), start=1):
response['transcript_' + str(i)] = answer
response['fixed_text_' + str(i)] = fixed
else:
response['transcript'] = text_answers[0]
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Answer(s) graded: {response}')
self._log(task, request_id, 'Requesting fixed text.')
response['fixed_text'] = await self._get_speaking_corrections(text_answers[0])
self._log(task, request_id, f'Fixed text: {response["fixed_text"]}')
if task in {1, 3}:
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - Adding perfect answer(s) to response.')
response['perfect_answer'] = perfect_answers[0]["answer"]
# TODO: check if it is answer["answer"] instead
for i, answer in enumerate(perfect_answers, start=1):
response['perfect_answer_' + str(i)] = answer
solutions = []
for file_name in temp_files:
solutions.append(await self._file_storage.upload_file_firebase_get_url(f'{FilePaths.FIREBASE_SPEAKING_VIDEO_FILES_PATH}{uuid4()}.wav', file_name))
self._logger.info(
f'POST - speaking_task_{task} - {request_id} - Adding transcript and fixed texts to response.'
)
response["overall"] = self._fix_speaking_overall(response["overall"], response["task_response"])
response["solutions"] = solutions
if task in {1,3}:
response["answer"] = solutions
else:
response["fullPath"] = solutions[0]
for i, answer in enumerate(text_answers, start=1):
response['transcript_' + str(i)] = answer
response['fixed_text_' + str(i)] = await self._get_speaking_corrections(answer)
else:
response['transcript'] = answers[0]["answer"]
self._log(task, request_id, f'Final response: {response}')
return response
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Requesting fixed text.')
response['fixed_text'] = await self._get_speaking_corrections(answers[0]["answer"])
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Fixed text: {response["fixed_text"]}')
finally:
for file_path in temp_files:
try:
if os.path.exists(file_path):
os.remove(file_path)
except Exception as e:
self._log(task, request_id, f'Error cleaning up temp file {file_path}: {str(e)}')
response['perfect_answer'] = perfect_answers[0]["answer"]
def _log(self, task: int, request_id: str, message: str):
self._logger.info(f'POST - speaking_task_{task} - {request_id} - {message}')
response["overall"] = self._fix_speaking_overall(response["overall"], response["task_response"])
self._logger.info(f'POST - speaking_task_{task} - {request_id} - Final response: {response}')
return response
@staticmethod
async def save_file(item: GradeSpeakingItem) -> str:
sound_file_name = "tmp/" + str(uuid.uuid4())
content = await item.answer.read()
async with aiofiles.open(sound_file_name, 'wb') as f:
await f.write(content)
return sound_file_name
# ==================================================================================================================
# grade_speaking_task helpers
@@ -336,7 +355,7 @@ class SpeakingService(ISpeakingService):
{
"role": "user",
"content": (
'For pronunciations act as if you heard the answers and they were transcripted '
'For pronunciations act as if you heard the answers and they were transcribed '
'as you heard them.'
)
},

29
app/services/impl/exam/writing.py
View File

@@ -1,3 +1,4 @@
import asyncio
from typing import List, Dict
from app.services.abc import IWritingService, ILLMService, IAIDetectorService
@@ -126,7 +127,7 @@ class WritingService(IWritingService):
TemperatureSettings.GEN_QUESTION_TEMPERATURE
)
response = await self._llm.prediction(
evaluation_promise = self._llm.prediction(
llm_model,
messages,
["comment"],
@@ -134,15 +135,27 @@ class WritingService(IWritingService):
)
perfect_answer_minimum = 150 if task == 1 else 250
perfect_answer = await self._get_perfect_answer(question, perfect_answer_minimum)
perfect_answer_promise = self._get_perfect_answer(question, perfect_answer_minimum)
fixed_text_promise = self._get_fixed_text(answer)
ai_detection_promise = self._ai_detector.run_detection(answer)
response["perfect_answer"] = perfect_answer["perfect_answer"]
response["overall"] = ExercisesHelper.fix_writing_overall(response["overall"], response["task_response"])
response['fixed_text'] = await self._get_fixed_text(answer)
prediction_result, perfect_answer_result, fixed_text_result, ai_detection_result = await asyncio.gather(
evaluation_promise,
perfect_answer_promise,
fixed_text_promise,
ai_detection_promise
)
ai_detection = await self._ai_detector.run_detection(answer)
if ai_detection is not None:
response['ai_detection'] = ai_detection
response = prediction_result
response["perfect_answer"] = perfect_answer_result["perfect_answer"]
response["overall"] = ExercisesHelper.fix_writing_overall(
response["overall"],
response["task_response"]
)
response['fixed_text'] = fixed_text_result
if ai_detection_result is not None:
response['ai_detection'] = ai_detection_result
return response

72
app/services/impl/third_parties/whisper.py
View File

@@ -1,22 +1,66 @@
import os
from fastapi.concurrency import run_in_threadpool
import threading
import whisper
import asyncio
from concurrent.futures import ThreadPoolExecutor
from typing import Dict
from whisper import Whisper
from app.services.abc import ISpeechToTextService
"""
The whisper model is not thread safe, a thread pool
with 4 whisper models will be created so it can
process up to 4 transcriptions at a time.
The base model requires ~1GB so 4 instances is the safe bet:
https://github.com/openai/whisper?tab=readme-ov-file#available-models-and-languages
"""
class OpenAIWhisper(ISpeechToTextService):
def __init__(self, model_name: str = "base", num_models: int = 4):
self._model_name = model_name
self._num_models = num_models
self._models: Dict[int, 'Whisper'] = {}
self._lock = threading.Lock()
self._next_model_id = 0
self._is_closed = False
def __init__(self, model: Whisper):
self._model = model
for i in range(num_models):
self._models[i] = whisper.load_model(self._model_name, in_memory=True)
async def speech_to_text(self, file_path):
if os.path.exists(file_path):
result = await run_in_threadpool(
self._model.transcribe, file_path, fp16=False, language='English', verbose=False
)
return result["text"]
else:
print("File not found:", file_path)
raise Exception("File " + file_path + " not found.")
self._executor = ThreadPoolExecutor(
max_workers=num_models,
thread_name_prefix="whisper_worker"
)
def get_model(self) -> 'Whisper':
with self._lock:
model_id = self._next_model_id
self._next_model_id = (self._next_model_id + 1) % self._num_models
return self._models[model_id]
async def speech_to_text(self, file_path: str) -> str:
if not os.path.exists(file_path):
raise FileNotFoundError(f"File {file_path} not found.")
def transcribe():
model = self.get_model()
return model.transcribe(
file_path,
fp16=False,
language='English',
verbose=False
)["text"]
loop = asyncio.get_running_loop()
return await loop.run_in_executor(self._executor, transcribe)
def close(self):
with self._lock:
if not self._is_closed:
self._is_closed = True
if self._executor:
self._executor.shutdown(wait=True, cancel_futures=True)
def __del__(self):
self.close()