import os
import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from langchain_core.prompts import PromptTemplate
from langchain_community.document_loaders import PyPDFLoader
from langchain.chains.question_answering import load_qa_chain
from langchain.llms import HuggingFacePipeline

# Load Mistral model
model_path = "nvidia/Mistral-NeMo-Minitron-8B-Base"
mistral_tokenizer = AutoTokenizer.from_pretrained(model_path)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
dtype = torch.bfloat16
mistral_model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=dtype, device_map=device)

# Create a pipeline for text generation
text_generation = pipeline(
    "text-generation",
    model=mistral_model,
    tokenizer=mistral_tokenizer,
    max_length=512,
    do_sample=True,
    temperature=0.7,
)

# Create a HuggingFacePipeline to use with langchain
mistral_pipeline = HuggingFacePipeline(pipeline=text_generation)

def initialize(file_path, question):
    try:
        if os.path.exists(file_path):
            # Load and split the PDF
            pdf_loader = PyPDFLoader(file_path)
            pages = pdf_loader.load_and_split()
            
            # Extract context from the first 30 pages (or all if less than 30)
            context = "\n".join(str(page.page_content) for page in pages[:30])
            
            # Create a prompt template
            prompt_template = """Answer the question as precisely as possible using the provided context. If the answer is
                                 not contained in the context, say "answer not available in context" \n\n
                                 Context: \n {context}?\n
                                 Question: \n {question} \n
                                 Answer:
                               """
            prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"])
            
            # Create a QA chain using the Mistral model
            qa_chain = load_qa_chain(mistral_pipeline, chain_type="stuff", prompt=prompt)
            
            # Run the chain
            result = qa_chain({"input_documents": pages, "question": question}, return_only_outputs=True)
            
            return result['output_text']
        else:
            return "Error: Unable to process the document. Please ensure the PDF file is valid."
    except Exception as e:
        return f"An error occurred: {str(e)}"

# Define Gradio Interface
input_file = gr.File(label="Upload PDF File")
input_question = gr.Textbox(label="Ask about the document")
output_text = gr.Textbox(label="Answer from Mistral")

def pdf_qa(file, question):
    if file is None:
        return "Please upload a PDF file first."
    return initialize(file.name, question)

# Create Gradio Interface
gr.Interface(
    fn=pdf_qa,
    inputs=[input_file, input_question],
    outputs=output_text,
    title="PDF Question Answering using Mistral Model",
    description="Upload a PDF file and ask questions about its content."
).launch(share=True)  # Added share=True here