mcq-deberta-v3-large

mcq-deberta-v3-large

Fine-tuned DeBERTa-v3-large for 5-option Multiple Choice Question answering.

Model Details

• Base: microsoft/deberta-v3-large
• Task: Text classification (num_labels=1, scores each option independently)
• Training: K-Fold cross-validation, MAP@3 metric
• Project: IIT Madras BS in Data Science — DL & GenAI (T2-2026)
• Author: Shitanshu Chaurasiya · Roll No. 24F2006167

Usage

from transformers import DebertaV2Tokenizer, DebertaV2ForSequenceClassification
import torch

tokenizer = DebertaV2Tokenizer.from_pretrained("Shitanshu06/mcq-deberta-v3-large")
model = DebertaV2ForSequenceClassification.from_pretrained("Shitanshu06/mcq-deberta-v3-large", num_labels=1)

question = "What is dropout?"
option   = "To prevent overfitting"

enc = tokenizer(question, option, return_tensors="pt", truncation=True, max_length=192)
with torch.no_grad():
    score = model(**enc).logits.item()

Live Demo

https://deberta-v3-large.onrender.com