math-misunderstanding-ettin-v1

Math Misunderstanding Classifier (Ettin-Encoder)

This model is fine-tuned to identify student math misconceptions. It was developed for the Eedi - Mining Misconceptions in Mathematics Kaggle competition.

Model Description

• Developed by: usmanqamr
• Base Model: jhu-clsp/ettin-encoder-400m (ModernBERT architecture)
• Number of Classes: 65 Misconception labels
• CV Score: 0.9428

Performance

The model was trained for 3 epochs and achieves a high Mean Average Precision (MAP@3) in detecting common student errors in geometry, algebra, and arithmetic.

How to Use

You can use this model directly with the Hugging Face transformers library:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "usmanqamr/math-misunderstanding-ettin-v1"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "Question: What is 1/2 + 1/3? Student Answer: 2/5"
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    logits = model(**inputs).logits
    
predicted_class = torch.argmax(logits, dim=-1)
print(predicted_class)