AventIQ-AI / pythia-410m-chatbot

Pythia Quantized Model for Question/Answer

This repository hosts a quantized version of the Pythia model, fine-tuned for question/answer tasks. The model has been optimized for efficient deployment while maintaining high accuracy, making it suitable for resource-constrained environments.

Model Details

• Model Architecture: Pythia-410m
• Task: Chatbot
• Dataset: sewon/ambig_qa
• Quantization: Float16
• Fine-tuning Framework: Hugging Face Transformers

Usage

Installation

pip install transformers torch

Loading the Model

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("EleutherAI/pythia-410m")
model = AutoModelForCausalLM.from_pretrained("AventIQ-AI/pythia-410m-chatbot")

tokenizer.pad_token = tokenizer.eos_token

def chat_with_model(model, tokenizer, question, max_length=256):
    """Generate response to a question"""
    input_text = question
    
    inputs = tokenizer(input_text, return_tensors="pt", padding=True, truncation=True, max_length=512)
    
    with torch.no_grad():
        outputs = model.generate(
            inputs["input_ids"],
            attention_mask=inputs["attention_mask"],  
            max_length=max_length,
            num_return_sequences=1,
            temperature=1.0,
            do_sample=True,  
            pad_token_id=tokenizer.pad_token_id
        )

    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Example usage
test_question = "What is the capital of France?"
response = chat_with_model(model, tokenizer, test_question)
print("Answer", response)

Performance Metrics

• Accuracy: 0.56
• F1 Score: 0.56
• Precision: 0.68
• Recall: 0.56

Fine-Tuning Details

Dataset

The Hugging Face's ambig_qa dataset was used, containing both question and answer examples.

Training

• Number of epochs: 3
• Batch size: 4
• Evaluation strategy: epoch
• Learning rate: 2e-5

Quantization

Post-training quantization was applied using PyTorch's built-in quantization framework to reduce the model size and improve inference efficiency.

Repository Structure

.
├── model/               # Contains the quantized model files
├── tokenizer/           # Tokenizer configuration and vocabulary files
├── model.safensors/     # Fine Tuned Model
├── README.md            # Model documentation

Limitations

• The model may not generalize well to domains outside the fine-tuning dataset.
• Quantization may result in minor accuracy degradation compared to full-precision models.

Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request if you have suggestions or improvements.