from fastapi import FastAPI, HTTPException, Header, Depends, BackgroundTasks, Query from fastapi.middleware.cors import CORSMiddleware from fastapi.responses import StreamingResponse from pydantic import BaseModel, Field from typing import List, Optional, Dict, AsyncGenerator import json import os import logging from txtai.embeddings import Embeddings import pandas as pd import glob import uuid import httpx import asyncio # Set up logging logging.basicConfig(level=logging.INFO) logger = logging.getLogger(__name__) app = FastAPI( title="Embeddings API", description="An API for creating and querying text embeddings indexes.", version="1.0.0" ) CHAT_AUTH_KEY = os.environ.get("CHAT_AUTH_KEY", "default_secret_key") # Enable CORS app.add_middleware( CORSMiddleware, allow_origins=["*"], # Allows all origins allow_credentials=True, allow_methods=["*"], # Allows all methods allow_headers=["*"], # Allows all headers ) embeddings = Embeddings({"path": "avsolatorio/GIST-all-MiniLM-L6-v2"}) class DocumentRequest(BaseModel): index_id: str = Field(..., description="Unique identifier for the index") documents: List[str] = Field(..., description="List of documents to be indexed") class QueryRequest(BaseModel): index_id: str = Field(..., description="Unique identifier for the index to query") query: str = Field(..., description="The search query") num_results: int = Field(..., description="Number of results to return", ge=1) def save_embeddings(index_id: str, document_list: List[str]): try: folder_path = f"/app/indexes/{index_id}" os.makedirs(folder_path, exist_ok=True) # Save embeddings embeddings.save(f"{folder_path}/embeddings") # Save document_list with open(f"{folder_path}/document_list.json", "w") as f: json.dump(document_list, f) logger.info(f"Embeddings and document list saved for index_id: {index_id}") except Exception as e: logger.error(f"Error saving embeddings for index_id {index_id}: {str(e)}") raise HTTPException(status_code=500, detail=f"Error saving embeddings: {str(e)}") def load_embeddings(index_id: str) -> List[str]: try: folder_path = f"/app/indexes/{index_id}" if not os.path.exists(folder_path): logger.error(f"Index not found for index_id: {index_id}") raise HTTPException(status_code=404, detail="Index not found") # Load embeddings embeddings.load(f"{folder_path}/embeddings") # Load document_list with open(f"{folder_path}/document_list.json", "r") as f: document_list = json.load(f) logger.info(f"Embeddings and document list loaded for index_id: {index_id}") return document_list except Exception as e: logger.error(f"Error loading embeddings for index_id {index_id}: {str(e)}") raise HTTPException(status_code=500, detail=f"Error loading embeddings: {str(e)}") @app.post("/create_index/", response_model=dict, tags=["Index Operations"]) async def create_index(request: DocumentRequest): """ Create a new index with the given documents. - **index_id**: Unique identifier for the index - **documents**: List of documents to be indexed """ try: document_list = [(i, text, None) for i, text in enumerate(request.documents)] embeddings.index(document_list) save_embeddings(request.index_id, request.documents) # Save the original documents logger.info(f"Index created successfully for index_id: {request.index_id}") return {"message": "Index created successfully"} except Exception as e: logger.error(f"Error creating index: {str(e)}") raise HTTPException(status_code=500, detail=f"Error creating index: {str(e)}") @app.post("/query_index/", response_model=dict, tags=["Index Operations"]) async def query_index(request: QueryRequest): """ Query an existing index with the given search query. - **index_id**: Unique identifier for the index to query - **query**: The search query - **num_results**: Number of results to return """ try: document_list = load_embeddings(request.index_id) results = embeddings.search(request.query, request.num_results) queried_texts = [document_list[idx[0]] for idx in results] logger.info(f"Query executed successfully for index_id: {request.index_id}") return {"queried_texts": queried_texts} except Exception as e: logger.error(f"Error querying index: {str(e)}") raise HTTPException(status_code=500, detail=f"Error querying index: {str(e)}") def process_csv_file(file_path): try: df = pd.read_csv(file_path) df_rows = df.apply(lambda row: ' '.join(row.values.astype(str)), axis=1) txtai_data = [(i, row, None) for i, row in enumerate(df_rows)] return txtai_data, df_rows.tolist() except Exception as e: logger.error(f"Error processing CSV file {file_path}: {str(e)}") return None, None def check_and_index_csv_files(): index_data_folder = "/app/index_data" if not os.path.exists(index_data_folder): logger.warning(f"index_data folder not found: {index_data_folder}") return csv_files = glob.glob(os.path.join(index_data_folder, "*.csv")) for csv_file in csv_files: index_id = os.path.splitext(os.path.basename(csv_file))[0] if not os.path.exists(f"/app/indexes/{index_id}"): logger.info(f"Processing CSV file: {csv_file}") txtai_data, documents = process_csv_file(csv_file) if txtai_data and documents: embeddings.index(txtai_data) save_embeddings(index_id, documents) logger.info(f"CSV file indexed successfully: {csv_file}") else: logger.warning(f"Failed to process CSV file: {csv_file}") else: logger.info(f"Index already exists for: {csv_file}") # ... [Previous code for DocumentRequest, QueryRequest, save_embeddings, load_embeddings, create_index, query_index, process_csv_file, check_and_index_csv_files remains the same] class ChatRequest(BaseModel): query: str = Field(..., description="The user's query") index_id: str = Field(..., description="Unique identifier for the index to query") conversation_id: Optional[str] = Field(None, description="Unique identifier for the conversation") model_id: str = Field(..., description="Identifier for the LLM model to use") user_id: str = Field(..., description="Unique identifier for the user") enable_followup: bool = Field(default=False, description="Flag to enable follow-up questions") async def get_api_key(x_api_key: str = Header(...)) -> str: if x_api_key != CHAT_AUTH_KEY: raise HTTPException(status_code=403, detail="Invalid API key") return x_api_key async def stream_llm_request(api_key: str, llm_request: Dict[str, str], endpoint_url:str), -> AsyncGenerator[str, None]: """ Make a streaming request to the LLM service. """ try: async with httpx.AsyncClient() as client: async with client.stream( "POST", endpoint_url, headers={ "accept": "text/event-stream", "X-API-Key": api_key, "Content-Type": "application/json" }, json=llm_request ) as response: if response.status_code != 200: raise HTTPException(status_code=response.status_code, detail="Error from LLM service") async for chunk in response.aiter_text(): yield chunk except httpx.HTTPError as e: logger.error(f"HTTP error occurred while making LLM request: {str(e)}") raise HTTPException(status_code=500, detail=f"HTTP error occurred while making LLM request: {str(e)}") except Exception as e: logger.error(f"Unexpected error occurred while making LLM request: {str(e)}") raise HTTPException(status_code=500, detail=f"Unexpected error occurred while making LLM request: {str(e)}") @app.post("/chat/", response_class=StreamingResponse, tags=["Chat"]) async def chat(request: ChatRequest, background_tasks: BackgroundTasks, api_key: str = Depends(get_api_key)): """ Chat endpoint that uses embeddings search and LLM for response generation. """ try: # Load embeddings for the specified index document_list = load_embeddings(request.index_id) # Perform embeddings search search_results = embeddings.search(request.query, 5) # Get top 5 relevant results context = "\n".join([document_list[idx[0]] for idx in search_results]) # Create RAG prompt rag_prompt = f"Based on the following context, please answer the user's question:\n\nContext:\n{context}\n\nUser's question: {request.query}\n\nAnswer:" system_prompt = "You are a helpful assistant tasked with providing answers using the context provided" # Generate conversation_id if not provided conversation_id = request.conversation_id or str(uuid.uuid4()) if request.enable_followup: # Prepare the request for the LLM service pass llm_request = { "query": rag_prompt, "model_id": 'openai/gpt-4o-mini', "conversation_id": conversation_id, "user_id": request.user_id endpoint_url = "https://pvanand-general-chat.hf.space/v2/followup-agent" else: llm_request = { "prompt": rag_prompt, "system_message": system_prompt, "model_id": request.model_id, "conversation_id": conversation_id, "user_id": request.user_id } endpoint_url = "https://pvanand-audio-chat.hf.space/llm-agent" logger.info(f"Starting chat response generation for user: {request.user_id} Full request: {llm_request}") async def response_generator(): full_response = "" async for chunk in stream_llm_request(api_key, llm_request,endpoint_url): full_response += chunk yield chunk logger.info(f"Finished chat response generation for user: {request.user_id} Full response{full_response}") # Here you might want to add logic to save the conversation or perform other background tasks # For example: # background_tasks.add_task(save_conversation, request.user_id, conversation_id, request.query, full_response) return StreamingResponse(response_generator(), media_type="text/event-stream") except Exception as e: logger.error(f"Error in chat endpoint: {str(e)}") raise HTTPException(status_code=500, detail=f"Error in chat endpoint: {str(e)}") @app.on_event("startup") async def startup_event(): check_and_index_csv_files() if __name__ == "__main__": import uvicorn uvicorn.run(app, host="0.0.0.0", port=7860)