app.py

from huggingface_hub import login
from fastapi import FastAPI, Depends, HTTPException
import logging
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModel
from services.qdrant_searcher import QdrantSearcher
from services.openai_service import generate_rag_response
from utils.auth import token_required
from dotenv import load_dotenv
import os
import torch

# Load environment variables from .env file
load_dotenv()

# Initialize FastAPI application
app = FastAPI()

# Set the cache directory for Hugging Face
os.environ["HF_HOME"] = "/tmp/huggingface_cache"

# Ensure the cache directory exists
hf_home_dir = os.environ["HF_HOME"]
if not os.path.exists(hf_home_dir):
    os.makedirs(hf_home_dir)

# Setup logging using Python's standard logging library
logging.basicConfig(level=logging.INFO)

# Load Hugging Face token from environment variable
huggingface_token = os.getenv('HUGGINGFACE_HUB_TOKEN')
if huggingface_token:
    try:
        login(token=huggingface_token, add_to_git_credential=True)
        logging.info("Successfully logged into Hugging Face Hub.")
    except Exception as e:
        logging.error(f"Failed to log into Hugging Face Hub: {e}")
        raise HTTPException(status_code=500, detail="Failed to log into Hugging Face Hub.")
else:
    raise ValueError("Hugging Face token is not set. Please set the HUGGINGFACE_HUB_TOKEN environment variable.")

# Initialize the Qdrant searcher
qdrant_url = os.getenv('QDRANT_URL')
access_token = os.getenv('QDRANT_ACCESS_TOKEN')

if not qdrant_url or not access_token:
    raise ValueError("Qdrant URL or Access Token is not set. Please set the QDRANT_URL and QDRANT_ACCESS_TOKEN environment variables.")

# Load the model and tokenizer with trust_remote_code=True
try:
    cache_folder = os.path.join(hf_home_dir, "transformers_cache")
    
    # Load the tokenizer and model with trust_remote_code=True
    tokenizer = AutoTokenizer.from_pretrained('nomic-ai/nomic-embed-text-v1.5', trust_remote_code=True)
    model = AutoModel.from_pretrained('nomic-ai/nomic-embed-text-v1.5', trust_remote_code=True)

    logging.info("Successfully loaded the model and tokenizer with transformers.")
    
    # Initialize the Qdrant searcher after the model is successfully loaded
    global searcher  # Ensure searcher is accessible globally if needed
    searcher = QdrantSearcher(qdrant_url=qdrant_url, access_token=access_token)

except Exception as e:
    logging.error(f"Failed to load the model or initialize searcher: {e}")
    raise HTTPException(status_code=500, detail="Failed to load the custom model or initialize searcher.")

# Function to embed text using the model
def embed_text(text):
    inputs = tokenizer(text, padding=True, truncation=True, return_tensors="pt")
    outputs = model(**inputs)
    embeddings = outputs.last_hidden_state.mean(dim=1)  # Example: mean pooling
    return embeddings.detach().numpy()

# Define the request body models
class SearchDocumentsRequest(BaseModel):
    query: str
    limit: int = 3

class GenerateRAGRequest(BaseModel):
    search_query: str

# Define the search documents endpoint
@app.post("/api/search-documents")
async def search_documents(
    body: SearchDocumentsRequest,
    credentials: tuple = Depends(token_required)
):
    customer_id, user_id = credentials

    if not customer_id or not user_id:
        logging.error("Failed to extract customer_id or user_id from the JWT token.")
        raise HTTPException(status_code=401, detail="Invalid token: missing customer_id or user_id")

    logging.info("Received request to search documents")
    try:
        logging.info("Starting document search")

        # Encode the query using the custom embedding function
        query_embedding = embed_text(body.query)

        collection_name = "my_embeddings"  # Use the collection name where the embeddings are stored    

        # Perform search using the precomputed embeddings
        hits, error = searcher.search_documents(collection_name, query_embedding, user_id, body.limit)
        
        if error:
            logging.error(f"Search documents error: {error}")
            raise HTTPException(status_code=500, detail=error)

        return hits
    except Exception as e:
        logging.error(f"Unexpected error: {e}")
        raise HTTPException(status_code=500, detail=str(e))

# Define the generate RAG response endpoint
@app.post("/api/generate-rag-response")
async def generate_rag_response_api(
    body: GenerateRAGRequest,
    credentials: tuple = Depends(token_required)
):
    customer_id, user_id = credentials

    if not customer_id or not user_id:
        logging.error("Failed to extract customer_id or user_id from the JWT token.")
        raise HTTPException(status_code=401, detail="Invalid token: missing customer_id or user_id")

    logging.info("Received request to generate RAG response")
    try:
        logging.info("Starting document search")

        # Encode the query using the custom embedding function
        query_embedding = embed_text(body.search_query)

        # Perform search using the precomputed embeddings
        hits, error = searcher.search_documents("documents", query_embedding, user_id)
        
        if error:
            logging.error(f"Search documents error: {error}")
            raise HTTPException(status_code=500, detail=error)

        logging.info("Generating RAG response")

        # Generate the RAG response using the retrieved documents
        response, error = generate_rag_response(hits, body.search_query)
        
        if error:
            logging.error(f"Generate RAG response error: {error}")
            raise HTTPException(status_code=500, detail=error)

        return {"response": response}
    except Exception as e:
        logging.error(f"Unexpected error: {e}")
        raise HTTPException(status_code=500, detail=str(e))

if __name__ == '__main__':
    import uvicorn
    uvicorn.run(app, host='0.0.0.0', port=8000)