app.py

import gradio as gr
from transformers import pipeline

# Set up the baseline QA model.
# This model is pre-finetuned on SQuAD and is known to provide reasonable answers.
model_id = "distilbert-base-uncased-distilled-squad"

# Create a question-answering pipeline.
# Set device=0 if you want to use a GPU, or remove it (or set device=-1) to run on CPU.
qa_pipeline = pipeline("question-answering", model=model_id, tokenizer=model_id, device=0)

def answer_question(question, context, max_seq_length, doc_stride, max_answer_length, n_best_size):
    """
    This function uses the baseline QA pipeline to extract an answer from the provided context.
    Hyperparameters:
      - max_seq_length: Maximum tokens for the combined input.
      - doc_stride: Overlap between segments when the context is too long.
      - max_answer_length: Maximum tokens in the extracted answer.
      - n_best_size: Number of candidate answer spans considered.
    """
    # Run the pipeline with the provided hyperparameters.
    result = qa_pipeline(
        question=question,
        context=context,
        max_length=int(max_seq_length),
        doc_stride=int(doc_stride),
        max_answer_length=int(max_answer_length),
        n_best_size=int(n_best_size)
    )
    return result.get("answer", "No answer found.")

# Define the Gradio interface.
interface = gr.Interface(
    fn=answer_question,
    inputs=[
        gr.Textbox(lines=2, placeholder="Enter your question here", label="Question"),
        gr.Textbox(lines=10, placeholder="Enter the context here", label="Context"),
        gr.Slider(minimum=100, maximum=512, value=384, step=1, label="Max Seq Length"),
        gr.Slider(minimum=50, maximum=256, value=128, step=1, label="Doc Stride"),
        gr.Slider(minimum=10, maximum=100, value=30, step=1, label="Max Answer Length"),
        gr.Slider(minimum=1, maximum=50, value=20, step=1, label="N Best Size")
    ],
    outputs=gr.Textbox(label="Answer"),
    title="Baseline QA Model Demo",
    description=(
        "This demo uses the baseline model (distilbert-base-uncased-distilled-squad) for question answering. "
        "Enter a question and a context, and adjust the hyperparameters as needed."
    )
)

# Launch the interface. When deployed in your HF Space, this app will automatically load.
if __name__ == "__main__":
    interface.launch()