research_buddy_app.py

from llama_index import Document
from llama_index.chat_engine import CondenseQuestionChatEngine
from llama_index.indices.vector_store import VectorIndexRetriever
from llama_index.node_parser import SimpleNodeParser
from llama_index import LangchainEmbedding, ServiceContext
from llama_index import VectorStoreIndex
from llama_index import StorageContext, load_index_from_storage
from llama_index.query_engine import RetrieverQueryEngine
from llama_index.response_synthesizers import TreeSummarize,get_response_synthesizer
from llama_index.llms import ChatMessage

from langchain.llms import Clarifai
from langchain.embeddings import ClarifaiEmbeddings


from clarifai_grpc.channel.clarifai_channel import ClarifaiChannel
from clarifai_grpc.grpc.api import resources_pb2, service_pb2, service_pb2_grpc
from clarifai_grpc.grpc.api.status import status_code_pb2


import uuid

import streamlit as st

import modal

CLARIFAI_PAT = st.secrets.CLARIFAI_PAT
MODERATION_THRESHOLD = st.secrets.MODERATION_THRESHOLD
st.set_page_config(page_title="Research Buddy: Insights and Q&A on AI Research Papers using GPT and Nougat", page_icon="🧐", layout="centered", initial_sidebar_state="auto", menu_items=None)
st.title(body="AI Research Buddy: Nougat + GPT Powered Paper Insights 📚🤖")
st.info("""This Application currently only works with arxiv and acl anthology web links which belong to the format:- 
1) Arxiv:- https://arxiv.org/abs/paper_unique_identifier
2) ACL Anthology:- https://aclanthology.org/paper_unique_identifier/ 

This Application uses the recently released Meta Nougat Visual Transformer for processing Papers""", icon="ℹ️")
user_input = st.text_input("Enter the arxiv or acl anthology url of the paper", "https://aclanthology.org/2023.semeval-1.266/")


def initialize_session_state():
    if "vector_store" not in st.session_state:
        st.session_state.vector_store = None

    if "messages" not in st.session_state.keys():
        st.session_state.messages = [
            {"role": "assistant", "content": "Ask me a question about the research paper"}
        ]

    if "paper_content" not in st.session_state:
        st.session_state.paper_content = None

    if "paper_insights" not in st.session_state:
        st.session_state.paper_insights = None


initialize_session_state()


def get_paper_content(url: str) -> str:
    with st.spinner(text="Using Nougat(https://facebookresearch.github.io/nougat/) to read the paper contents and get the markdown representation of the paper"):
        f = modal.Function.lookup("streamlit-hack", "main")
        output = f.call(url)
        st.session_state.paper_content = output
        return output


def index_paper_content(content: str):
    with st.spinner(text="Indexing the paper – hang tight! This should take 3-5 minutes"):
        try:
            LLM_USER_ID = 'openai'
            LLM_APP_ID = 'chat-completion'
            # Change these to whatever model and text URL you want to use
            LLM_MODEL_ID = 'GPT-3_5-turbo'
            llm = Clarifai(pat=CLARIFAI_PAT, user_id=LLM_USER_ID, app_id=LLM_APP_ID, model_id=LLM_MODEL_ID)

            documents = [Document(text=content)]
            parser = SimpleNodeParser.from_defaults()

            nodes = parser.get_nodes_from_documents(documents)
            USER_ID = 'openai'
            APP_ID = 'embed'
            # Change these to whatever model and text URL you want to use
            MODEL_ID = 'text-embedding-ada'
            embeddings = ClarifaiEmbeddings(pat=CLARIFAI_PAT, user_id=USER_ID, app_id=APP_ID, model_id=MODEL_ID)
            embed_model = LangchainEmbedding(embeddings)
            service_context = ServiceContext.from_defaults(llm=llm, embed_model=embed_model)
            index = VectorStoreIndex(nodes, service_context=service_context)
            persist_dir = uuid.uuid4().hex
            st.session_state.vector_store = persist_dir
            index.storage_context.persist(persist_dir=persist_dir)
            return "Paper has been Indexed"

        except Exception as e:
            print(str(e))
            return "Unable to Index the Research Paper"


def generate_insights():
    with st.spinner(text="Generating insights on the paper and preparing the Chatbot"):
        try:
            LLM_USER_ID = 'openai'
            LLM_APP_ID = 'chat-completion'
            # Change these to whatever model and text URL you want to use
            LLM_MODEL_ID = 'GPT-3_5-turbo'
            llm = Clarifai(pat=CLARIFAI_PAT, user_id=LLM_USER_ID, app_id=LLM_APP_ID, model_id=LLM_MODEL_ID)

            USER_ID = 'openai'
            APP_ID = 'embed'
            # Change these to whatever model and text URL you want to use
            MODEL_ID = 'text-embedding-ada'
            embeddings = ClarifaiEmbeddings(pat=CLARIFAI_PAT, user_id=USER_ID, app_id=APP_ID, model_id=MODEL_ID)
            embed_model = LangchainEmbedding(embeddings)

            service_context = ServiceContext.from_defaults(llm=llm, embed_model=embed_model)

            index = load_index_from_storage(
                StorageContext.from_defaults(persist_dir=st.session_state.vector_store),
                service_context=service_context
            )

            retriever = VectorIndexRetriever(
                index=index,
                similarity_top_k=4,
            )
            # configure response synthesizer
            response_synthesizer = get_response_synthesizer(
                response_mode="tree_summarize", service_context=service_context
            )

            # assemble query engine
            query_engine = RetrieverQueryEngine(
                retriever=retriever,
                response_synthesizer=response_synthesizer,
            )

            response_key_insights = query_engine.query("Generate core crux insights, contributions and results of the paper as Key Topics and thier content in markdown format where each Key Topic is in bold followed by its content")

        except Exception as e:
            print(str(e))
            response_key_insights = "Error While Generating Insights"

        st.session_state.paper_insights = response_key_insights.response


if st.button("Read and Index Paper"):
    paper_content = get_paper_content(url=user_input)

    if st.session_state.paper_content is not None:
        with st.expander("See Paper Contents"):
            st.markdown(paper_content)

        result = index_paper_content(content=paper_content)
        st.write(result)
        generate_insights()


if st.session_state.paper_content is not None:
    with st.expander("See Paper Contents"):
        st.markdown(st.session_state.paper_content)

if st.session_state.paper_insights is not None:
    st.sidebar.title("# 🚀 Illuminating Research Insights 📜💡")
    st.sidebar.write(st.session_state.paper_insights)


def reset_conversation():
    st.session_state.messages = [
        {"role": "assistant", "content": "Ask me a question about the research paper"}
    ]


def moderate_text(text: str) -> tuple:
    MODERATION_USER_ID = 'clarifai'
    MODERATION_APP_ID = 'main'
    # Change these to whatever model and text URL you want to use
    MODERATION_MODEL_ID = 'moderation-multilingual-text-classification'
    MODERATION_MODEL_VERSION_ID = '79c2248564b0465bb96265e0c239352b'

    channel = ClarifaiChannel.get_grpc_channel()
    stub = service_pb2_grpc.V2Stub(channel)

    metadata = (('authorization', 'Key ' + CLARIFAI_PAT),)

    userDataObject = resources_pb2.UserAppIDSet(user_id=MODERATION_USER_ID, app_id=MODERATION_APP_ID)

    # To use a local text file, uncomment the following lines
    # with open(TEXT_FILE_LOCATION, "rb") as f:
    #    file_bytes = f.read()

    post_model_outputs_response = stub.PostModelOutputs(
        service_pb2.PostModelOutputsRequest(
            user_app_id=userDataObject,
            # The userDataObject is created in the overview and is required when using a PAT
            model_id=MODERATION_MODEL_ID,
            version_id=MODERATION_MODEL_VERSION_ID,  # This is optional. Defaults to the latest model version
            inputs=[
                resources_pb2.Input(
                    data=resources_pb2.Data(
                        text=resources_pb2.Text(
                            raw=text
                        )
                    )
                )
            ]
        ),
        metadata=metadata
    )
    if post_model_outputs_response.status.code != status_code_pb2.SUCCESS:
        print(post_model_outputs_response.status)
        raise Exception("Post model outputs failed, status: " + post_model_outputs_response.status.description)

    # Since we have one input, one output will exist here
    output = post_model_outputs_response.outputs[0]
    moderation_reasons = ""
    intervention_required = False
    for concept in output.data.concepts:
        if concept.value > MODERATION_THRESHOLD:
            moderation_reasons += concept.name + ","
            intervention_required = True

    return moderation_reasons, intervention_required


if st.session_state.vector_store is not None:
    LLM_USER_ID = 'openai'
    LLM_APP_ID = 'chat-completion'
    # Change these to whatever model and text URL you want to use
    LLM_MODEL_ID = 'GPT-3_5-turbo'
    llm = Clarifai(pat=CLARIFAI_PAT, user_id=LLM_USER_ID, app_id=LLM_APP_ID, model_id=LLM_MODEL_ID)

    USER_ID = 'openai'
    APP_ID = 'embed'
    # Change these to whatever model and text URL you want to use
    MODEL_ID = 'text-embedding-ada'
    embeddings = ClarifaiEmbeddings(pat=CLARIFAI_PAT, user_id=USER_ID, app_id=APP_ID, model_id=MODEL_ID)
    embed_model = LangchainEmbedding(embeddings)

    service_context = ServiceContext.from_defaults(llm=llm, embed_model=embed_model)

    index = load_index_from_storage(
        StorageContext.from_defaults(persist_dir=st.session_state.vector_store),
        service_context=service_context
    )

    retriever = VectorIndexRetriever(
        index=index,
        similarity_top_k=2,
    )
    # configure response synthesizer
    response_synthesizer = get_response_synthesizer(
        response_mode="tree_summarize", service_context=service_context
    )

    # assemble query engine
    query_engine = RetrieverQueryEngine(
        retriever=retriever,
        response_synthesizer=response_synthesizer,
    )

    custom_chat_history = []
    for message in st.session_state.messages:
        custom_message = ChatMessage(role=message["role"], content=message["content"])
        custom_chat_history.append(custom_message)

    chat_engine = CondenseQuestionChatEngine.from_defaults(service_context=service_context, query_engine=query_engine,
                                                           verbose=True,
                                                           chat_history=custom_chat_history)

    if prompt := st.chat_input("Your question"):  # Prompt for user input and save to chat history
        st.session_state.messages.append({"role": "user", "content": prompt})

    st.button('Reset Chat', on_click=reset_conversation)

    for message in st.session_state.messages:  # Display the prior chat messages
        with st.chat_message(message["role"]):
            st.write(message["content"])

    # If last message is not from assistant, generate a new response
    if st.session_state.messages[-1]["role"] != "assistant":
        with st.chat_message("assistant"):
            with st.spinner("Thinking..."):
                try:
                    reason, intervene = moderate_text(prompt)
                except Exception as e:
                    print(str(e))
                    reason = ''
                    intervene = False
                if not intervene:
                    response = chat_engine.chat(prompt)
                    st.write(response.response)
                    message = {"role": "assistant", "content": response.response}
                    st.session_state.messages.append(message)  # Add response to message history
                else:
                    response = f"This query cannot be processed as it has been detected to be {reason}"
                    st.write(response)
                    message = {"role": "assistant", "content": response}
                    st.session_state.messages.append(message)