app.py

import os
import random
import gradio as gr
from langchain_community.document_loaders import PyPDFLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.chains import RetrievalQA
from langchain_groq import ChatGroq
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough


# Initialize the FAISS vector store
vector_store = None

# Sample PDF file
sample_filename = "Attention Is All You Need.pdf"

examples_questions = [["What is Transformer?"],
            ["What is Attention?"],
            ["What is Scaled Dot-Product Attention?"],
            ["What are Encoder and Decoder?"],
            ["Describe more about the Transformer."],
            ["Why use self-attention?"],
           ]

template = \
"""Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say that you don't know, don't try to make up an answer.
Always say "Thanks for asking!" at the end of the answer.

{context}

Question: {question}

Answer:
"""

# Function to handle PDF upload and indexing
def index_pdf(pdf):
    global vector_store
    
    # Load the PDF
    loader = PyPDFLoader(pdf.name)
    documents = loader.load()

    # Split the documents into chunks
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    texts = text_splitter.split_documents(documents)

    # Embed the chunks 
    embeddings = HuggingFaceEmbeddings(model_name="bert-base-uncased", encode_kwargs={"normalize_embeddings": True})

    # Store the embeddings in the vector store
    vector_store = FAISS.from_documents(texts, embeddings)

    return "PDF indexed successfully!"

def load_sample_pdf():
    global vector_store
    
    # Load the PDF
    loader = PyPDFLoader(sample_filename)
    documents = loader.load()

    # Split the documents into chunks
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    texts = text_splitter.split_documents(documents)

    # Embed the chunks 
    embeddings = HuggingFaceEmbeddings(model_name="bert-base-uncased", encode_kwargs={"normalize_embeddings": True})

    # Store the embeddings in the vector store
    vector_store = FAISS.from_documents(texts, embeddings)

    return "Sample PDF indexed successfully!"


def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)
    

def generate_response(query, history, model, temperature, max_tokens, top_p, seed):
    if vector_store is None:
        return "Please upload and index a PDF at the Indexing tab."

    if seed == 0:
        seed = random.randint(1, 100000)

    retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 16})
    llm = ChatGroq(groq_api_key=os.environ.get("GROQ_API_KEY"), model=model)
    custom_rag_prompt = PromptTemplate.from_template(template)
   
    rag_chain = (
        {"context": retriever | format_docs, "question": RunnablePassthrough()}
        | custom_rag_prompt
        | llm
        | StrOutputParser()
    )

    response = rag_chain.invoke(query)
   
    return response



additional_inputs = [
    gr.Dropdown(choices=["llama-3.1-70b-versatile", "llama-3.1-8b-instant", "llama3-70b-8192", "llama3-8b-8192", "mixtral-8x7b-32768", "gemma2-9b-it", "gemma-7b-it"], value="llama-3.1-70b-versatile", label="Model"),
    gr.Slider(minimum=0.0, maximum=1.0, step=0.01, value=0.5, label="Temperature", info="Controls diversity of the generated text. Lower is more deterministic, higher is more creative."),
    gr.Slider(minimum=1, maximum=8000, step=1, value=8000, label="Max Tokens", info="The maximum number of tokens that the model can process in a single response.<br>Maximums: 8k for gemma 7b it, gemma2 9b it, llama 7b & 70b, 32k for mixtral 8x7b, 132k for llama 3.1."),
    gr.Slider(minimum=0.0, maximum=1.0, step=0.01, value=0.5, label="Top P", info="A method of text generation where a model will only consider the most probable next tokens that make up the probability p."),
    gr.Number(precision=0, value=0, label="Seed", info="A starting point to initiate generation, use 0 for random")
]


def greet(name):
    return f"Hello, {name}!"



with gr.Blocks(theme="Nymbo/Alyx_Theme") as demo:
    # Set the title and description for the app.
    gr.Markdown("# Simple Gradio Greeter")
    gr.Markdown("Enter your name and get a personalized greeting!")

    # Define the input component (a text box for the name).
    name_input = gr.Textbox(
        lines=2,
        placeholder="Enter your name here...",
        label="Your Name"
    )

    # Define the output component (a text box for the greeting).
    greeting_output = gr.Textbox(label="Greeting")

    # Add a button that will trigger the 'greet' function when clicked.
    # The 'fn' argument specifies the function to call.
    # The 'inputs' argument specifies which component's value to pass to 'fn'.
    # The 'outputs' argument specifies which component will display the return value of 'fn'.
    submit_button = gr.Button("Get Greeting")
    submit_button.click(
        fn=greet,
        inputs=name_input,
        outputs=greeting_output
    )

# Launch the Gradio app.
# The `share=True` argument creates a public, shareable link (useful for demos).
# In a local environment, it will also open a local URL.
demo.launch()