import os import time from fastapi import FastAPI,Request from fastapi.responses import HTMLResponse from fastapi.staticfiles import StaticFiles from llama_index.core import StorageContext, load_index_from_storage, VectorStoreIndex, SimpleDirectoryReader, ChatPromptTemplate, Settings from llama_index.llms.huggingface import HuggingFaceInferenceAPI from llama_index.embeddings.huggingface import HuggingFaceEmbedding from pydantic import BaseModel from fastapi.responses import JSONResponse import uuid # for generating unique IDs import datetime from fastapi.middleware.cors import CORSMiddleware from fastapi.templating import Jinja2Templates from huggingface_hub import InferenceClient import json import re # Define Pydantic model for incoming request body class MessageRequest(BaseModel): message: str repo_id = "meta-llama/Meta-Llama-3-8B-Instruct" llm_client = InferenceClient( model=repo_id, token=os.getenv("HF_TOKEN"), ) def summarize_conversation(inference_client: InferenceClient, history: list): # Construct the full prompt with history history_text = "\n".join([f"{entry['sender']}: {entry['message']}" for entry in history]) full_prompt = f"{history_text}\n\nSummarize the conversation in three concise points only give me only Summarization in python list formate :\n" response = inference_client.post( json={ "inputs": full_prompt, "parameters": {"max_new_tokens": 512}, "task": "text-generation", }, ) # Decode the response generated_text = json.loads(response.decode())[0]["generated_text"] # Use regex to extract the list inside brackets matches = re.findall(r'\[(.*?)\]', generated_text) # If matches found, extract the content if matches: # Assuming we only want the first match, split by commas and strip whitespace list_items = matches[0].split(',') cleaned_list = [item.strip() for item in list_items] return cleaned_list else: return generated_text os.environ["HF_TOKEN"] = os.getenv("HF_TOKEN") app = FastAPI() @app.middleware("http") async def add_security_headers(request: Request, call_next): response = await call_next(request) response.headers["Content-Security-Policy"] = "frame-ancestors *; frame-src *; object-src *;" response.headers["X-Frame-Options"] = "ALLOWALL" return response # Allow CORS requests from any domain app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"], ) @app.get("/favicon.ico") async def favicon(): return HTMLResponse("") # or serve a real favicon if you have one app.mount("/static", StaticFiles(directory="static"), name="static") templates = Jinja2Templates(directory="static") # Configure Llama index settings Settings.llm = HuggingFaceInferenceAPI( model_name="meta-llama/Meta-Llama-3-8B-Instruct", tokenizer_name="meta-llama/Meta-Llama-3-8B-Instruct", context_window=3000, token=os.getenv("HF_TOKEN"), max_new_tokens=512, generate_kwargs={"temperature": 0.1}, ) Settings.embed_model = HuggingFaceEmbedding( model_name="BAAI/bge-small-en-v1.5" ) PERSIST_DIR = "db" PDF_DIRECTORY = 'data' # Ensure directories exist os.makedirs(PDF_DIRECTORY, exist_ok=True) os.makedirs(PERSIST_DIR, exist_ok=True) chat_history = [] current_chat_history = [] def data_ingestion_from_directory(): documents = SimpleDirectoryReader(PDF_DIRECTORY).load_data() storage_context = StorageContext.from_defaults() index = VectorStoreIndex.from_documents(documents) index.storage_context.persist(persist_dir=PERSIST_DIR) def initialize(): start_time = time.time() data_ingestion_from_directory() # Process PDF ingestion at startup print(f"Data ingestion time: {time.time() - start_time} seconds") initialize() # Run initialization tasks def handle_query(query): chat_text_qa_msgs = [ ( "user", """ You are the Clara Redfernstech chatbot. Your goal is to provide accurate, professional, and helpful answers to user queries based on the company's data. Always ensure your responses are clear and concise. Give response within 10-15 words only {context_str} Question: {query_str} """ ) ] text_qa_template = ChatPromptTemplate.from_messages(chat_text_qa_msgs) storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR) index = load_index_from_storage(storage_context) context_str = "" for past_query, response in reversed(current_chat_history): if past_query.strip(): context_str += f"User asked: '{past_query}'\nBot answered: '{response}'\n" query_engine = index.as_query_engine(text_qa_template=text_qa_template, context_str=context_str) answer = query_engine.query(query) if hasattr(answer, 'response'): response=answer.response elif isinstance(answer, dict) and 'response' in answer: response =answer['response'] else: response ="Sorry, I couldn't find an answer." current_chat_history.append((query, response)) return response @app.get("/ch/{id}", response_class=HTMLResponse) async def load_chat(request: Request, id: str): return templates.TemplateResponse("index.html", {"request": request, "user_id": id}) # Route to save chat history @app.post("/hist/") async def save_chat_history(history: dict): # Logic to save chat history, using the `id` from the frontend print(history) # You can replace this with actual save logic cleaned_summary = summarize_conversation(llm_client, history) print(cleaned_summary) return {"message": "Chat history saved"} @app.post("/webhook") async def receive_form_data(request: Request): form_data = await request.json() # Generate a unique ID (for tracking user) unique_id = str(uuid.uuid4()) # Here you can do something with form_data like saving it to a database print("Received form data:", form_data) # Send back the unique id to the frontend return JSONResponse({"id": unique_id}) @app.post("/chat/") async def chat(request: MessageRequest): message = request.message # Access the message from the request body response = handle_query(message) # Process the message message_data = { "sender": "User", "message": message, "response": response, "timestamp": datetime.datetime.now().isoformat() } chat_history.append(message_data) return {"response": response} @app.get("/") def read_root(): return {"message": "Welcome to the API"}