---
library_name: transformers
tags:
  - text-generation
  - llama
  - mcq-generation
  - distractor-generation
  - safetensors
---

# LLaMA 3.2 1B - Distractors Generation Model

This model is designed to generate distractors (incorrect but plausible answer choices) for multiple-choice questions. It is particularly useful in educational applications where high-quality distractors are needed.

## Model Details

- **Model Name**: `llama3.2_1B_distractors_generation`
- **Architecture**: LLaMA 3.2 1B
- **Developer**: [BirendraSharma](https://huggingface.co/BirendraSharma)
- **Use Case**: MCQ distractor generation
- **License**: [More Information Needed]

## How to Use the Model

You can load the model and tokenizer using `transformers` and generate distractors using the following example:

### Installation
Ensure you have `transformers` and `torch` installed:
```bash
pip install transformers torch
```

### Loading the Model & Tokenizer
```python
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
import torch

# Load model and tokenizer
model_path = "BirendraSharma/llama3.2_1B_distractors_generation"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16, device_map="auto")

def generate_distractors(context, question, answer, instruction, model, tokenizer, max_seq_length=1024):
    """Generates distractors for a given question-answer pair."""
    model.eval()
    prompt = f"""{instruction}
    Context: '{context}'
    Question: '{question}'
    Answer: '{answer}'.
    Provide only three distractors as a comma-separated list. Do not include explanations, commentary, or additional text."""
    
    inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True, max_length=max_seq_length).to("cuda")
    
    generation_config = GenerationConfig(
        max_new_tokens=128,
        use_cache=True,
        temperature=0.7,
    )
    
    outputs = model.generate(**inputs, generation_config=generation_config)
    distractors_text = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0][len(prompt):].strip()
    distractors = [d.strip() for d in distractors_text.split(",") if d.strip()]
    
    return ", ".join(distractors[:3])

# Example usage
context = "In physics, the concept of jerk is used to describe the rate of change of acceleration."
question = "What is the physical quantity that jerk is the rate of change of?"
answer = "Acceleration"
instruction = "Generate plausible but incorrect answer choices."

distractors = generate_distractors(context, question, answer, instruction, model, tokenizer)
print("Distractors:", distractors)


# Output: velocity, momentum, force
```

## Dataset Used for Training
[More Information Needed]

## Training Details
- **Hardware Used**: T4 GPU (Google Colab)
- **Fine-tuned from**: LLaMA 3.2 1B
- **Training Framework**: `transformers` + `SFTTrainer`

## Evaluation Metrics
The model was evaluated using BLEU and ROUGE scores to compare generated distractors with reference distractors.

## Citation
If you use this model, please cite:
```
@article{llama3.2_distractor_gen,
  title={Distractor Generation using LLaMA 3.2},
  author={Birendra Sharma},
  year={2025},
  publisher={Hugging Face Model Hub}
}
```

---

For more details, visit the [Hugging Face model page](https://huggingface.co/BirendraSharma/llama3.2_1B_distractors_generation).