QA Trainer 📚🤖

QA Trainer is a Python package for fine-tuning Question Answering models using feedback data.
It uses SimpleTransformers for model training and Optuna for hyperparameter tuning.

📂 Example Dataset Format

The dataset must be a JSON array of feedback entries. Each entry should have:

• "question" → The question text
• "answer" → The answer text (used as both context and answer in training)
• "feedback" → A positive number to include it in training, negative/zero to exclude

Example feedback_data.json for question answer model but format will remain same for other model also you can take help of simple transformers python libary

[
    {
        "question": "What is AI?",
        "answer": "AI stands for Artificial Intelligence.",
        "feedback": 1
    },
    {
        "question": "Who is the president of Mars?",
        "answer": "Elon Musk.",
        "feedback": -1
    },
    {
        "question": "What is Python?",
        "answer": "Python is a programming language.",
        "feedback": 1
    }
]

🚀 Quick Start

from feedback_rl_transformers import train_with_feedback

train_with_feedback(
    feedback_data_path="feedback_data.json",  # Path to your feedback dataset
    model_type="bert",                        # Model type (see table below)
    model_path="bert-base-uncased",           # Pretrained model name or path
    output_dir="fine_tuned_model",            # Where to save the final model
    n_trials=5,                               # Optuna hyperparameter tuning trials
    use_cuda=False                            # Set True to use GPU if available
)

🧠 Supported Models

*Some large models may require more memory or a GPU.

⚙️ Parameters

📤 Output

After training:

• The fine-tuned model will be saved in output_dir
• Best hyperparameters from Optuna will be printed