app.py

import gradio as gr
import uuid
from typing import Sequence

from langchain.chains import create_history_aware_retriever, create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_community.document_loaders import TextLoader
from langchain_core.messages import AIMessage, BaseMessage, HumanMessage
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.vectorstores import InMemoryVectorStore

from langchain_groq import ChatGroq
from langchain_huggingface import HuggingFaceEmbeddings

from langchain_text_splitters import RecursiveCharacterTextSplitter
from langgraph.checkpoint.memory import MemorySaver
from langgraph.graph import START, StateGraph
from langgraph.graph.message import add_messages
from typing_extensions import Annotated, TypedDict

import os
GROQ_API_KEY = os.getenv("GROQ_API_KEY")

llm = ChatGroq(model="llama-3.2-11b-text-preview", api_key=GROQ_API_KEY, temperature=0)

loader = TextLoader("stj.txt")
loader.load()
docs = loader.load()

model_name = "sentence-transformers/all-MiniLM-L6-v2"
model_kwargs = {'device': 'cpu'}
encode_kwargs = {'normalize_embeddings': False}
hf = HuggingFaceEmbeddings(
    model_name=model_name,
    model_kwargs=model_kwargs,
    encode_kwargs=encode_kwargs
)

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
splits = text_splitter.split_documents(docs)
vectorstore = InMemoryVectorStore.from_documents(
    documents=splits, embedding=hf
)
retriever = vectorstore.as_retriever()

contextualize_q_system_prompt = (
    "Sohbet geçmişi ve en son kullanıcı sorusu verilirse, sohbet geçmişine atıfta bulunabilecek en son kullanıcı sorusunu, sohbet geçmişi olmadan anlaşılabilecek bağımsız bir soru haline getirin. Soruyu yanıtlamayın, sadece yeniden düzenleyin ve gerekirse geri döndürün."
)
contextualize_q_prompt = ChatPromptTemplate.from_messages(
    [
        ("system", contextualize_q_system_prompt),
        MessagesPlaceholder("chat_history"),
        ("human", "{input}"),
    ]
)
history_aware_retriever = create_history_aware_retriever(
    llm, retriever, contextualize_q_prompt
)

system_prompt = (
    "Soru-cevap görevleri için bir asistansın. Soruyu yanıtlamak için alınan aşağıdaki bağlam parçalarını kullan. Cevabı bilmiyorsan, bilmiyorum de. Cevabı üç cümleyle sınırla ve kısa tut."
    "\n\n"
    "{context}"
)
qa_prompt = ChatPromptTemplate.from_messages(
    [
        ("system", system_prompt),
        MessagesPlaceholder("chat_history"),
        ("human", "{input}"),
    ]
)
question_answer_chain = create_stuff_documents_chain(llm, qa_prompt)

rag_chain = create_retrieval_chain(history_aware_retriever, question_answer_chain)

class State(TypedDict):
    input: str
    chat_history: Annotated[Sequence[BaseMessage], add_messages]
    context: str
    answer: str

def call_model(state: State):
    response = rag_chain.invoke(state)
    return {
        "chat_history": [
            HumanMessage(state["input"]),
            AIMessage(response["answer"]),
        ],
        "context": response["context"],
        "answer": response["answer"],
    }

workflow = StateGraph(state_schema=State)
workflow.add_edge(START, "model")
workflow.add_node("model", call_model)

memory = MemorySaver()
app = workflow.compile(checkpointer=memory)

def rag_response(user_input, chat_history, thread_id):
    config = {"configurable": {"thread_id": thread_id}}

    state = {
        "input": user_input,
        "chat_history": [msg[0] for msg in chat_history]
    }

    result = app.invoke(state, config=config)

    chat_history.append((user_input, result["answer"]))

    return "", chat_history

def acknowledge_disclaimer():
    thread_id = str(uuid.uuid4())
    return gr.update(visible=False), gr.update(visible=False), thread_id

with gr.Blocks() as demo:
    
    disclaimer_message = gr.Markdown(
        "**⚠️ Uyarı:** Büyük dil modelleri [halüsinasyon](https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)) sebebi ile yanlış cevaplar verebilir. Lütfen aldığınız cevapları uygulamadan önce doğrulayınız.\nBu demo resmi olarak Işık Üniversitesini temsil **etmemektedir.**",
        visible=True
    )
    ok_button = gr.Button("OK", visible=True)
    
    chatbox = gr.Chatbot(label="Sohbet Geçmişi", visible=False)
    user_input = gr.Textbox(placeholder="Soru", label="Kullanıcı Soru Alanı", visible=False)
    submit_button = gr.Button("Submit", visible=False)
    
    thread_id_component = gr.State()

    ok_button.click(acknowledge_disclaimer, outputs=[disclaimer_message, ok_button, thread_id_component])

    submit_button.click(rag_response, inputs=[user_input, chatbox, thread_id_component], outputs=[user_input, chatbox])

    ok_button.click(lambda: [gr.update(visible=True)] * 3, outputs=[chatbox, user_input, submit_button])

demo.launch()