A fine-tuned language model that generates high-quality physics multiple-choice questions for university entrance exam preparation with customizable cognitive skill levels based on Bloom's Taxonomy.

Model Details

Model Description

This model is specifically designed to generate competitive physics multiple-choice questions with accurate content, plausible distractors, and appropriate difficulty levels for entrance exam preparation. It supports four cognitive skill levels (Recall, Application, Analysis, Evaluation) and excels across major physics domains including mechanics, electromagnetism, thermodynamics, optics, and modern physics.

Developed by: [flanara]
Model type: Fine-tuned Causal Language Model
Language(s) (NLP): English
License: MIT
Finetuned from model: microsoft/phi-2

Model Sources

Repository: https://huggingface.co/flanara/physics-mcq-generator

Uses

Direct Use

This model is intended for direct use in generating physics multiple-choice questions for:

University entrance exam preparation with varying cognitive levels
Differentiated instruction materials
Bloom's Taxonomy-aligned assessment creation
Educational content creation across cognitive domains
Tutoring and teaching assistance with skill-based questioning

Downstream Use

The model can be integrated into:

Educational platforms with adaptive learning paths
Automated question bank generators with cognitive level filtering
Physics tutoring applications with skill-based progression
Exam preparation software with customized difficulty curves
Teacher tools for creating balanced assessments

Out-of-Scope Use

Generating questions for high-stakes exams without expert validation
Creating medical or safety-critical content
Replacing human physics educators entirely
Generating content outside physics domain
Using for psychological or cognitive assessment

Bias, Risks, and Limitations

Limitations

Performance is best on classical physics topics; may struggle with advanced quantum mechanics
Generated questions should always be reviewed by subject matter experts
Limited context length (~512 tokens) may affect complex question generation
Training data primarily from international curriculum standards
Cognitive skill differentiation may not be perfect for all topics

Risks

Potential for generating incorrect physics concepts if prompted unusually
May reflect biases present in the training data
Should not be used for high-stakes assessment without human oversight
Cognitive level assignments may not always match intended complexity

Recommendations

Users should:

Always verify generated questions with physics experts
Use as a tool to assist educators, not replace them
Disclose AI-generated content when used in educational materials
Monitor and review outputs for accuracy and appropriateness
Validate cognitive skill level assignments for important assessments

How to Get Started with the Model

Basic Usage with Cognitive Skills

python

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

# Load the model
model = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-2",
    device_map="auto",
    torch_dtype=torch.float16,
    trust_remote_code=True
)
model = PeftModel.from_pretrained(model, "your_username/physics-mcq-generator")
tokenizer = AutoTokenizer.from_pretrained("your_username/physics-mcq-generator")
tokenizer.pad_token = tokenizer.eos_token

def generate_physics_mcq(chapter, topic, difficulty="Medium", cognitive_skill="Application"):
    """
    Generate a physics MCQ with customizable cognitive skill level
    
    Cognitive Skill Levels:
    - 'Recall': Basic fact recall and definition questions
    - 'Application': Applying concepts to solve problems
    - 'Analysis': Analyzing situations and relationships
    - 'Evaluation': Complex reasoning and critical evaluation
    """
    
    prompt = f"""### Instruction:
Generate a multiple-choice question (MCQ) for a university entrance exam in Physics.

### Input:
Subject: Physics | Chapter: {chapter} | Topic: {topic} | Difficulty: {difficulty} | Cognitive_Skill: {cognitive_skill}

### Response:
Question:"""
    
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=250,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
    
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Examples with different cognitive skills
print("🧠 Recall Question (Basic knowledge):")
mcq = generate_physics_mcq("Mechanics", "Newton's Laws", "Easy", "Recall")
print(mcq)

print("\n⚡ Application Question (Problem solving):")
mcq = generate_physics_mcq("Electromagnetism", "Ohm's Law", "Medium", "Application")
print(mcq)

print("\n🔍 Analysis Question (Complex reasoning):")
mcq = generate_physics_mcq("Thermodynamics", "First Law", "Hard", "Analysis")
print(mcq)

print("\n🎯 Evaluation Question (Critical thinking):")
mcq = generate_physics_mcq("Modern Physics", "Quantum Mechanics", "Hard", "Evaluation")
print(mcq)

flanara/physics-mcq-generator

Physics MCQ Generator