Brushed up the backend, added writing task 1 academic prompt gen and grading ENCOA-274
This commit is contained in:
Carlos-Mesquita
207
ielts_be/services/impl/exam/writing/grade.py
Normal file
207
ielts_be/services/impl/exam/writing/grade.py
Normal file
@@ -0,0 +1,207 @@
|
||||
import asyncio
|
||||
from typing import Dict, Optional
|
||||
from uuid import uuid4
|
||||
|
||||
from ielts_be.configs.constants import GPTModels, TemperatureSettings
|
||||
from ielts_be.helpers import TextHelper, ExercisesHelper, FileHelper
|
||||
from ielts_be.repositories import IFileStorage
|
||||
from ielts_be.services import ILLMService, IAIDetectorService
|
||||
|
||||
|
||||
class GradeWriting:
|
||||
|
||||
def __init__(self, llm: ILLMService, file_storage: IFileStorage, ai_detector: IAIDetectorService):
|
||||
self._llm = llm
|
||||
self._file_storage = file_storage
|
||||
self._ai_detector = ai_detector
|
||||
|
||||
async def grade_writing_task(self, task: int, question: str, answer: str, attachment: Optional[str] = None):
|
||||
bare_minimum = 100 if task == 1 else 180
|
||||
|
||||
if not TextHelper.has_words(answer):
|
||||
return self._zero_rating("The answer does not contain enough english words.")
|
||||
elif not TextHelper.has_x_words(answer, bare_minimum):
|
||||
return self._zero_rating("The answer is insufficient and too small to be graded.")
|
||||
else:
|
||||
template = self._get_writing_template()
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": (
|
||||
f'You are a helpful assistant designed to output JSON on this format: {template}'
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
f'Evaluate the given Writing Task {task} response based on the IELTS grading system, '
|
||||
'ensuring a strict assessment that penalizes errors. Deduct points for deviations '
|
||||
'from the task, and assign a score of 0 if the response fails to address the question. '
|
||||
'Additionally, provide a detailed commentary highlighting both strengths and '
|
||||
'weaknesses in the response. '
|
||||
f'\n Question: "{question}" \n Answer: "{answer}"')
|
||||
}
|
||||
]
|
||||
|
||||
if task == 1:
|
||||
if attachment is None:
|
||||
messages.append({
|
||||
"role": "user",
|
||||
"content": (
|
||||
'Refer to the parts of the letter as: "Greeting Opener", "bullet 1", "bullet 2", '
|
||||
'"bullet 3", "closer (restate the purpose of the letter)", "closing greeting"'
|
||||
)
|
||||
})
|
||||
else:
|
||||
uuid = str(uuid4())
|
||||
name = attachment.split('/')[-1]
|
||||
out_path = f'./tmp/{uuid}/{name}'
|
||||
path = await self._file_storage.download_firebase_file(attachment, out_path)
|
||||
messages.append({
|
||||
"type": "image_url",
|
||||
"image_url": {
|
||||
"url": f"data:image/{name.split('.')[-1]};base64,{FileHelper.encode_image(path)}"
|
||||
}
|
||||
})
|
||||
|
||||
llm_model = GPTModels.GPT_3_5_TURBO if task == 1 else GPTModels.GPT_4_O
|
||||
temperature = (
|
||||
TemperatureSettings.GRADING_TEMPERATURE
|
||||
if task == 1 else
|
||||
TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
evaluation_promise = self._llm.prediction(
|
||||
llm_model,
|
||||
messages,
|
||||
["comment"],
|
||||
temperature
|
||||
)
|
||||
|
||||
perfect_answer_minimum = 150 if task == 1 else 250
|
||||
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)
|
||||
|
||||
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
|
||||
)
|
||||
|
||||
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
|
||||
|
||||
async def _get_fixed_text(self, text):
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": (
|
||||
'You are a helpful assistant designed to output JSON on this format: '
|
||||
'{"fixed_text": "fixed test with no misspelling errors"}'
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
'Fix the errors in the given text and put it in a JSON. '
|
||||
f'Do not complete the answer, only replace what is wrong. \n The text: "{text}"'
|
||||
)
|
||||
}
|
||||
]
|
||||
|
||||
response = await self._llm.prediction(
|
||||
GPTModels.GPT_3_5_TURBO,
|
||||
messages,
|
||||
["fixed_text"],
|
||||
0.2,
|
||||
False
|
||||
)
|
||||
return response["fixed_text"]
|
||||
|
||||
async def _get_perfect_answer(self, question: str, size: int) -> Dict:
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": (
|
||||
'You are a helpful assistant designed to output JSON on this format: '
|
||||
'{"perfect_answer": "perfect answer for the question"}'
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": f'Write a perfect answer for this writing exercise of a IELTS exam. Question: {question}'
|
||||
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": f'The answer must have at least {size} words'
|
||||
}
|
||||
]
|
||||
return await self._llm.prediction(
|
||||
GPTModels.GPT_4_O,
|
||||
messages,
|
||||
["perfect_answer"],
|
||||
TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def _zero_rating(comment: str):
|
||||
return {
|
||||
'comment': comment,
|
||||
'overall': 0,
|
||||
'task_response': {
|
||||
'Task Achievement': {
|
||||
"grade": 0.0,
|
||||
"comment": ""
|
||||
},
|
||||
'Coherence and Cohesion': {
|
||||
"grade": 0.0,
|
||||
"comment": ""
|
||||
},
|
||||
'Lexical Resource': {
|
||||
"grade": 0.0,
|
||||
"comment": ""
|
||||
},
|
||||
'Grammatical Range and Accuracy': {
|
||||
"grade": 0.0,
|
||||
"comment": ""
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@staticmethod
|
||||
def _get_writing_template():
|
||||
return {
|
||||
"comment": "comment about student's response quality",
|
||||
"overall": 0.0,
|
||||
"task_response": {
|
||||
"Task Achievement": {
|
||||
"grade": 0.0,
|
||||
"comment": "comment about Task Achievement of the student's response"
|
||||
},
|
||||
"Coherence and Cohesion": {
|
||||
"grade": 0.0,
|
||||
"comment": "comment about Coherence and Cohesion of the student's response"
|
||||
},
|
||||
"Lexical Resource": {
|
||||
"grade": 0.0,
|
||||
"comment": "comment about Lexical Resource of the student's response"
|
||||
},
|
||||
"Grammatical Range and Accuracy": {
|
||||
"grade": 0.0,
|
||||
"comment": "comment about Grammatical Range and Accuracy of the student's response"
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user