model/Engessay_grading_ML.py

from typing import *
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from .common import Grader

model_name = "KevSun/Engessay_grading_ML"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)


class Engessay_grading_ML(Grader):
    def info(self) -> str:
        return "[KevSun/Engessay_grading_ML](https://huggingface.co/KevSun/Engessay_grading_ML)"

    @torch.no_grad()
    def grade(self, question: str, answer: str) -> Tuple[float, str]:
        text = f"{question} {answer}"

        inputs = tokenizer(text, return_tensors="pt")

        outputs = model(**inputs)
        predictions = outputs.logits.squeeze()

        predicted_scores = predictions.numpy()
        scaled_scores = 2.25 * predicted_scores - 1.25
        rounded_scores = [round(score * 2) / 2 for score in scaled_scores]

        labels = [
            "cohesion",
            "syntax",
            "vocabulary",
            "phraseology",
            "grammar",
            "conventions",
        ]
        overall_score = round(sum(rounded_scores) / len(rounded_scores) * 2) / 2

        comment = ""
        for label, score in zip(labels, rounded_scores):
            comment += f"{label}: {score}\n"

        return overall_score, comment