app.py

import gradio as gr
import base64
import os
from openai import OpenAI

api_key = os.getenv('API_KEY')
base_url = os.getenv("BASE_URL")

client = OpenAI(
    api_key=api_key,
    base_url=base_url,
)


def extract_pdf_pypdf(pdf_dir):
    import fitz
    path = pdf_dir

    try:
        doc = fitz.open(path)
    except:
        print("can not read pdf")
        return None

    page_count = doc.page_count
    file_content = ""
    for page in range(page_count):
        text = doc.load_page(page).get_text("text")
        # 防止目录中包含References
        file_content += text + "\n\n"

    return file_content


def openai_api(messages):
    try:
        completion = client.chat.completions.create(
            model="claude-3-5-sonnet-20240620",
            messages=messages,
            temperature=0.1,
            max_tokens=8192,
            # timeout=300,
            stream=True
        )
    except Exception as ex:
        print("api 出现如下异常%s" % ex)
        return None

    if completion:
        try:
            response_2_list = [chunk.choices[0].delta.content if chunk.choices[0].delta.content else "" for chunk in
                               completion]
            print("response tokens:", len(response_2_list))

            response_2_content = ''.join(response_2_list)
            return response_2_content
        except Exception as ex:
            print("第二轮 出现如下异常%s" % ex)
            return None
    else:
        print("第二轮出现异常")
        return None


def predict(input_text, pdf_file):
    if pdf_file is None:
        return "Please upload a PDF file to proceed."

    file_content = extract_pdf_pypdf(pdf_file.name)
    messages = [
        {
            "role": "system",
            "content": "You are an expert in information extraction from scientific literature.",
        },
        {"role": "user", "content": """Provided Text:
    '''
    {{""" + file_content + """}}
    '''
                                        """ + input_text}
    ]
    extract_result = openai_api(messages)

    return extract_result or "Too many users. Please wait a moment!"


def view_pdf(pdf_file, max_pages=3):
    if pdf_file is None:
        return "Please upload a PDF file to view."

    try:
        # Open the PDF file
        doc = fitz.open(pdf_file.name)

        # Only read up to `max_pages` pages to reduce size for large PDFs
        preview_pdf = fitz.open()  # Create an empty PDF for the preview
        for page_num in range(min(max_pages, doc.page_count)):
            preview_pdf.insert_pdf(doc, from_page=page_num, to_page=page_num)

        # Save the preview as a temporary in-memory file
        pdf_data = preview_pdf.tobytes()

        # Encode as base64 for embedding in HTML
        b64_data = base64.b64encode(pdf_data).decode('utf-8')
        return f"<embed src='data:application/pdf;base64,{b64_data}' type='application/pdf' width='100%' height='700px' />"

    except Exception as e:
        print(f"Error displaying PDF: {e}")
        return "Error displaying PDF. Please try re-uploading."


en_1 = """Could you please help me extract the information of 'title'/'journal'/'year'/'author'/'institution'/'email' from the previous content in a markdown table format?
If any of this information was not available in the paper, please replace it with the string `""`. If the property contains multiple entities, please use a list to contain.
"""

en_2 = """Could you please help me extract the information of 'title'/'journal'/'year'/'author'/'institution'/'email' from the previous content in a JSON format?
If any of this information was not available in the paper, please replace it with the string `""`. If the property contains multiple entities, please use a list to contain.
"""

examples = [[en_1], [en_2]]

with gr.Blocks(title="PaperExtractGPT") as demo:
    gr.Markdown(
        '''<p align="center">
        <h1 align="center"> Paper Extract GPT </h1>
        <p> How to use:
        <br> <strong>1</strong>: Upload your PDF.
        <br> <strong>2</strong>: Click "View PDF" to preview it.
        <br> <strong>3</strong>: Enter your extraction prompt in the input box.
        <br> <strong>4</strong>: Click "Generate" to extract, and the extracted information will display below.
        </p>
        '''
    )
    with gr.Row():
        with gr.Column():
            gr.Markdown('## Upload PDF')
            file_input = gr.File(label="Upload your PDF", type="filepath")
            viewer_button = gr.Button("View PDF")
            file_out = gr.HTML(label="PDF Preview")

        with gr.Column():
            model_input = gr.Textbox(lines=7, placeholder='Enter your extraction prompt here', label='Input Prompt')
            example = gr.Examples(examples=examples, inputs=model_input)
            with gr.Row():
                gen = gr.Button("Generate")
                clr = gr.Button("Clear")
            outputs = gr.Markdown(label='Output', show_label=True,  value="""| Title                                       | Journal            | Year | Author                                        | Institution                                           | Email                 |
|---------------------------------------------|--------------------|------|-----------------------------------------------|-------------------------------------------------------|-----------------------|
| Paleomagnetic Study of Deccan Traps from Jabalpur to Amarkantak, Central India | J. Geomag. Geoelectr. | 1973 | R. K. VERMA, G. PULLAIAH, G.R. ANJANEYULU, P. K. MALLIK | National Geophysical Research Institute, Hyderabad, and Indian School o f Mines, Dhanbad | "" |
""")

    gen.click(fn=predict, inputs=[model_input, file_input], outputs=outputs)
    clr.click(fn=lambda: [gr.update(value=""), gr.update(value="")], inputs=None, outputs=[model_input, outputs])
    viewer_button.click(view_pdf, inputs=file_input, outputs=file_out)

demo.launch()