model/IELTS_essay_scoring.py

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

model_name = "JacobLinCool/IELTS_essay_scoring_safetensors"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)


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

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

        inputs = tokenizer(
            text, return_tensors="pt", padding=True, truncation=True, max_length=512
        )

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

        predicted_scores = predictions.numpy()
        normalized_scores = (predicted_scores / predicted_scores.max()) * 9
        rounded_scores = np.round(normalized_scores * 2) / 2

        labels = [
            "Task Achievement",
            "Coherence and Cohesion",
            "Vocabulary",
            "Grammar",
            "Overall",
        ]
        overall_score = float(rounded_scores[-1])

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

        return overall_score, comment