app.py

import os
import spaces
import gradio as gr
from huggingface_hub import InferenceClient
from pydantic import BaseModel

HF_TOKEN = os.getenv("HF_TOKEN")

class PaperAnalysis(BaseModel):
    title: str
    abstract_summary: str

response_format = {
    "type": "json_schema",
    "json_schema": {
        "name": "PaperAnalysis",
        "schema": PaperAnalysis.model_json_schema(),
        "strict": True,
    },
}

client = InferenceClient(
    provider="auto",
    api_key=HF_TOKEN,
)

@spaces.GPU(duration=60)
def extract_info(paper_text: str):
    if not paper_text.strip():
        return {"title": "", "abstract_summary": ""}
    messages = [
        {"role": "system", "content": "Extract the paper title and summarize its abstract."},
        {"role": "user", "content": paper_text},
    ]
    resp = client.chat.completions.create(
        model="meta-llama/Llama-3.2-1B-Instruct",
        messages=messages,
        response_format=response_format,
    )
    # parse response
    parsed = resp.choices[0].message
    # assuming the message is a dict
    return {
        "title": parsed.content.get("title", ""),
        "abstract_summary": parsed.content.get("abstract_summary", ""),
    }

with gr.Blocks() as demo:
    gr.Markdown("# 🎓 Paper Analysis Tool")
    with gr.Row():
        paper_input = gr.Textbox(label="Paper Text (include Title/Abstract)", lines=10)
        with gr.Column():
            title_out = gr.Textbox(label="Title", lines=1)
            summary_out = gr.Textbox(label="Abstract Summary", lines=5)
    analyze_btn = gr.Button("Extract Info")
    analyze_btn.click(fn=extract_info, inputs=paper_input, outputs=[title_out, summary_out])

if __name__ == "__main__":
    demo.launch()