app.py

# AUTOGENERATED! DO NOT EDIT! File to edit: ../drive/MyDrive/Codici/Python/Apps/Gradio_App/Langchain_apps/langchain_summarization_app.ipynb.

# %% auto 0
__all__ = ['hub_llm', 'title', 'description', 'combine_prompt_template', 'pdf_example_1', 'pdf_example_2', 'prompt_example_1',
           'prompt_example_2', 'upload_file_input', 'custom_prompt_input', 'custom_chunk_input', 'chunk_size_input',
           'chunk_overlap_input', 'examples', 'outputs', 'iface', 'summarize']

# %% ../drive/MyDrive/Codici/Python/Apps/Gradio_App/Langchain_apps/langchain_summarization_app.ipynb 3
from langchain_community.llms import HuggingFaceHub
from langchain_community.document_loaders import PyPDFLoader
from langchain_core.prompts import PromptTemplate
from langchain.chains import LLMChain
from langchain.text_splitter import CharacterTextSplitter
from langchain.chains.mapreduce import MapReduceChain
from langchain.prompts import PromptTemplate
from langchain.docstore.document import Document
from langchain.chains.summarize import load_summarize_chain

import os
import dotenv
from dotenv import load_dotenv
load_dotenv()

# # Ensure the API token is set 
# huggingfacehub_api_token = os.getenv('HUGGINGFACEHUB_API_TOKEN') 
# if not huggingfacehub_api_token: 
#   raise ValueError("Please set the HUGGINGFACEHUB_API_TOKEN environment variable.") 

# # %% ../drive/MyDrive/Codici/Python/Apps/Gradio_App/Langchain_apps/langchain_summarization_app.ipynb 5
# hub_llm = HuggingFaceHub(
#     repo_id="facebook/bart-large-cnn", # facebook/bart-large-cnn or "google/flan-t5-base" or "google/pegasus-xsum"
#     model_kwargs={
#         "temperature": 0.01, # Controls randomness (0.0: deterministic, 1.0: very random)
#         "max_new_tokens": 256*2,  # Maximum number of tokens to generate in the summary
#         "min_length": 30,  # Minimum length of the generated summary
#         "repetition_penalty": 1.2,  # Penalizes repeated tokens (higher value = less repetition)
#         "top_k": 50,  # Consider only the top k most likely tokens when generating
#         "top_p": 0.95,  # Consider tokens with cumulative probability up to top_p
#         "early_stopping": True, # Stops generation when a certain condition is met (e.g., end-of-sequence token)
#         "huggingfacehub_api_token": huggingfacehub_api_token
#     }
# )

# %% ../drive/MyDrive/Codici/Python/Apps/Gradio_App/Langchain_apps/langchain_summarization_app.ipynb 15
from langchain.text_splitter import RecursiveCharacterTextSplitter
import gradio as gr
import time

title="PDF Summarizer"
description="Summarize your PDF using a custom combine prompt."

# Default combine_prompt
combine_prompt_template = """Write a comprehensive summary of this academic article.

Divide the summary in:
1. Main Objective of the paper
2. Results

{text}

SUMMARY:"""

# Example PDF files and prompts
pdf_example_1 = './ZeroShotDataAug.pdf'
pdf_example_2 = './bert.pdf'
prompt_example_1 = """Write a comprehensive summary of this academic article.

Divide the summary in:
1. Main Objective of the paper
2. Results

{text}

SUMMARY:"""
prompt_example_2 = """Summarize the following document focusing on the key findings and methodology.

{text}

Summary:"""

# Initialize hub_llm outside the function, but without the token yet
hub_llm = HuggingFaceHub(
    repo_id="facebook/bart-large-cnn",
    model_kwargs={
        "temperature": 0.01,
        "max_new_tokens": 256 * 2,
        "min_length": 30,
        "repetition_penalty": 1.2,
        "top_k": 50,
        "top_p": 0.95,
        "early_stopping": True,
        # "huggingfacehub_api_token": huggingfacehub_api_token  # Add token later
    }
)

def summarize(pdf_file, custom_prompt, custom_chunk, chunk_size, chunk_overlap):
    try:
        # Load .env file and get the token
        load_dotenv()
        huggingfacehub_api_token = os.getenv('HUGGINGFACEHUB_API_TOKEN')

        if not huggingfacehub_api_token:
            raise ValueError("HUGGINGFACEHUB_API_TOKEN not found in .env")

        # Set the token for hub_llm within the function
        hub_llm.model_kwargs["huggingfacehub_api_token"] = huggingfacehub_api_token
    
        try:
            loader = PyPDFLoader(pdf_file.name)
            if custom_chunk:
                text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
                docs = loader.load_and_split(text_splitter=text_splitter)
            else:
                docs = loader.load_and_split()

            PROMPT = PromptTemplate(template=custom_prompt, input_variables=['text'])
            chain = load_summarize_chain(hub_llm, chain_type='map_reduce', combine_prompt=PROMPT)

            # Introduce a delay before calling the API
            time.sleep(1)
            summary = chain.run(docs)
            return summary
        except Exception as e:
            return f"An error occurred: {e}"
    
    except Exception as e:
        return f"An error occurred: {e}"

upload_file_input = gr.File(label="Upload PDF", type="filepath")
custom_prompt_input = gr.Textbox(label="Custom Prompt",
                                 lines=10,
                                 value=combine_prompt_template,
                                 info="Define your own prompt or leave empty for default.")
custom_chunk_input = gr.Checkbox(label="Custom Chunk", value=False, info="Recommended to be left unchecked")
chunk_size_input = gr.Number(label="Chunk Size", value=700,minimum=500,maximum=1000,step=100)
chunk_overlap_input = gr.Number(label="Chunk Overlap", value=50,minimum=10,maximum=100,step=100)

examples=[
    [pdf_example_1, prompt_example_1, False, 700, 50],
    # [pdf_example_2, prompt_example_2, False, 700, 50]
    ]

outputs = gr.Textbox(label="Summary")

iface = gr.Interface(
    title=title,
    description=description,
    fn=summarize,
    inputs=[upload_file_input,
            custom_prompt_input,
            custom_chunk_input,
            chunk_size_input,
            chunk_overlap_input
            ],
    outputs=outputs,
    examples=examples,
)

iface.launch(
    debug=False,
    # share=False,
    )