general_rag.py

import transformers
import pickle
import os
import re
import numpy as np
import torchvision
import nltk
import torch
import pandas as pd
import requests
import zipfile
import tempfile
from openai import OpenAI
from PyPDF2 import PdfReader
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import (
    AutoTokenizer,
    AutoModelForSeq2SeqLM,
    AutoModelForTokenClassification,
    AutoModelForCausalLM,
    pipeline, 
    Qwen2Tokenizer, 
    BartForConditionalGeneration
)
from sentence_transformers import SentenceTransformer, CrossEncoder, util
from sklearn.metrics.pairwise import cosine_similarity
from bs4 import BeautifulSoup
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file 
from typing import List, Dict, Optional
from safetensors.numpy import load_file
from safetensors.torch import safe_open
nltk.download('punkt_tab')


app = FastAPI()
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)
models = {}
data = {}

class QueryRequest(BaseModel):
    query: str
    language_code: int = 1

class ChatQuery(BaseModel):
    query: str
    language_code: int = 1
    #conversation_id: str

class ChatMessage(BaseModel):
    role: str
    content: str
    timestamp: str

def init_nltk():
    try:
        nltk.download('punkt', quiet=True)
        return True
    except Exception as e:
        print(f"Error initializing NLTK: {e}")
        return False


def get_completion(prompt: str, model: str = "deepseek/deepseek-prover-v2:free") -> str:
    api_key = os.environ.get('OPENROUTER_API_KEY')
    if not api_key:
        raise HTTPException(status_code=500, detail="OPENROUTER_API_KEY not found in environment variables")
    
    client = OpenAI(
        base_url="https://openrouter.ai/api/v1",
        api_key=api_key
    )
    
    if not prompt.strip():
        raise HTTPException(status_code=400, detail="Please enter a question")
    
    try:
        completion = client.chat.completions.create(
            extra_headers={
                "HTTP-Referer": "https://huggingface.co/spaces/thechaiexperiment/phitrial",
                "X-Title": "My Hugging Face Space"
            },
            model=model,
            messages=[
                {
                    "role": "user",
                    "content": prompt
                }
            ]
        )
        
        if (completion and 
            hasattr(completion, 'choices') and 
            completion.choices and 
            hasattr(completion.choices[0], 'message') and 
            hasattr(completion.choices[0].message, 'content')):
            return completion.choices[0].message.content
        else:
            raise HTTPException(status_code=500, detail="Received invalid response from API")
            
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

def load_models():
    try:
        print("Loading models...")
        device = "cuda" if torch.cuda.is_available() else "cpu"
        print(f"Device set to use {device}")
        models['embedding_model'] = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
        models['cross_encoder'] = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2', max_length=512)
        models['semantic_model'] = SentenceTransformer('all-MiniLM-L6-v2')
        models['ar_to_en_tokenizer'] = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
        models['ar_to_en_model'] = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
        models['en_to_ar_tokenizer'] = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ar")
        models['en_to_ar_model'] = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-en-ar")
        models['att_tokenizer'] = AutoTokenizer.from_pretrained("facebook/bart-base")
        models['att_model'] = BartForConditionalGeneration.from_pretrained("facebook/bart-base")
        models['bio_tokenizer'] = AutoTokenizer.from_pretrained("blaze999/Medical-NER")
        models['bio_model'] = AutoModelForTokenClassification.from_pretrained("blaze999/Medical-NER")
        models['ner_pipeline'] = pipeline("ner", model=models['bio_model'], tokenizer=models['bio_tokenizer'])
        model_name = "M4-ai/Orca-2.0-Tau-1.8B"
        models['llm_tokenizer'] = AutoTokenizer.from_pretrained(model_name)
        models['llm_model'] = AutoModelForCausalLM.from_pretrained(model_name)
        models['gen_tokenizer'] = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-1.7B-Instruct")
        models['gen_model'] = AutoModelForCausalLM.from_pretrained("HuggingFaceTB/SmolLM-1.7B-Instruct")
        print("Models loaded successfully")
        return True
    except Exception as e:
        print(f"Error loading models: {e}")
        return False

def load_embeddings() -> Optional[Dict[str, np.ndarray]]:
    try:
        embeddings_path = 'embeddings.safetensors'
        if not os.path.exists(embeddings_path):
            print("File not found locally. Attempting to download from Hugging Face Hub...")
            embeddings_path = hf_hub_download(
                repo_id=os.environ.get('HF_SPACE_ID', 'thechaiexperiment/TeaRAG'),
                filename="embeddings.safetensors",
                repo_type="space"
            )
        
        embeddings = {}
        with safe_open(embeddings_path, framework="pt") as f:
            keys = f.keys()
            for key in keys:
                try:
                    tensor = f.get_tensor(key)
                    if not isinstance(tensor, torch.Tensor):
                        raise TypeError(f"Value for key {key} is not a valid PyTorch tensor.")                 
                    embeddings[key] = tensor.numpy()
                except Exception as key_error:
                    print(f"Failed to process key {key}: {key_error}")
        if embeddings:
            print("Embeddings successfully loaded.")
        else:
            print("No embeddings could be loaded. Please check the file format and content.")        
        return embeddings
    except Exception as e:
        print(f"Error loading embeddings: {e}")
        return None

def normalize_key(key: str) -> str:
    match = re.search(r'file_(\d+)', key)
    if match:
        return match.group(1)  
    return key

def load_recipes_embeddings() -> Optional[np.ndarray]:
    try:
        embeddings_path = 'recipes_embeddings.safetensors'        
        if not os.path.exists(embeddings_path):
            print("File not found locally. Attempting to download from Hugging Face Hub...")
            embeddings_path = hf_hub_download(
                repo_id=os.environ.get('HF_SPACE_ID', 'thechaiexperiment/TeaRAG'),
                filename="embeddings.safetensors",
                repo_type="space"
            )
        embeddings = load_file(embeddings_path)
        if "embeddings" not in embeddings:
            raise ValueError("Key 'embeddings' not found in the safetensors file.")
        tensor = embeddings["embeddings"]      
        print(f"Successfully loaded embeddings.")
        print(f"Shape of embeddings: {tensor.shape}")
        print(f"Dtype of embeddings: {tensor.dtype}")
        print(f"First few values of the first embedding: {tensor[0][:5]}")
        return tensor
    except Exception as e:
        print(f"Error loading embeddings: {e}")
        return None

def load_documents_data(folder_path='downloaded_articles/downloaded_articles'):
    try:
        print("Loading documents data...")
        if not os.path.exists(folder_path) or not os.path.isdir(folder_path):
            print(f"Error: Folder '{folder_path}' not found")
            return False
        html_files = [f for f in os.listdir(folder_path) if f.endswith('.html')]
        if not html_files:
            print(f"No HTML files found in folder '{folder_path}'")
            return False
        documents = []
        for file_name in html_files:
            file_path = os.path.join(folder_path, file_name)
            try:
                with open(file_path, 'r', encoding='utf-8') as file:
                    soup = BeautifulSoup(file, 'html.parser')
                    text = soup.get_text(separator='\n').strip()
                    documents.append({"file_name": file_name, "content": text})
            except Exception as e:
                print(f"Error reading file {file_name}: {e}")
            data['df'] = pd.DataFrame(documents)        
            if data['df'].empty:
                print("No valid documents loaded.")
                return False
            print(f"Successfully loaded {len(data['df'])} document records.")
            return True
    except Exception as e:
        print(f"Error loading docs: {e}")
        return None    

def load_data():
    embeddings_success = load_embeddings()
    documents_success = load_documents_data()    
    if not embeddings_success:
        print("Warning: Failed to load embeddings, falling back to basic functionality")
    if not documents_success:
        print("Warning: Failed to load documents data, falling back to basic functionality")       
    return True

print("Initializing application...")
init_success = load_models() and load_data()


def translate_text(text, source_to_target='ar_to_en'):
    try:
        if source_to_target == 'ar_to_en':
            tokenizer = models['ar_to_en_tokenizer']
            model = models['ar_to_en_model']
        else:
            tokenizer = models['en_to_ar_tokenizer']
            model = models['en_to_ar_model']            
        inputs = tokenizer(text, return_tensors="pt", truncation=True)
        outputs = model.generate(**inputs)
        return tokenizer.decode(outputs[0], skip_special_tokens=True)
    except Exception as e:
        print(f"Translation error: {e}")
        return text

def embed_query_text(query_text):
    embedding = models['embedding_model']
    query_embedding = embedding.encode([query_text])
    return query_embedding

def query_embeddings(query_embedding, embeddings_data, n_results):
    embeddings_data = load_embeddings()
    if not embeddings_data:
        print("No embeddings data available.")
        return []
    try:
        doc_ids = list(embeddings_data.keys())
        doc_embeddings = np.array(list(embeddings_data.values()))
        similarities = cosine_similarity(query_embedding, doc_embeddings).flatten()
        top_indices = similarities.argsort()[-n_results:][::-1]
        return [(doc_ids[i], similarities[i]) for i in top_indices]
    except Exception as e:
        print(f"Error in query_embeddings: {e}")
        return []

def query_recipes_embeddings(query_embedding, embeddings_data, n_results):
    embeddings_data = load_recipes_embeddings()
    if embeddings_data is None:
        print("No embeddings data available.")
        return []
    try:
        if query_embedding.ndim == 1:
            query_embedding = query_embedding.reshape(1, -1)
        similarities = cosine_similarity(query_embedding, embeddings_data).flatten()
        top_indices = similarities.argsort()[-n_results:][::-1]
        return [(index, similarities[index]) for index in top_indices]
    except Exception as e:
        print(f"Error in query_recipes_embeddings: {e}")
        return []

def get_page_title(url):
    try:
        response = requests.get(url)
        if response.status_code == 200:
            soup = BeautifulSoup(response.text, 'html.parser')
            title = soup.find('title')
            return title.get_text() if title else "No title found"
        else:
            return None
    except requests.exceptions.RequestException:
        return None

def retrieve_document_texts(doc_ids, folder_path='downloaded_articles/downloaded_articles'):
    texts = []
    for doc_id in doc_ids:
        file_path = os.path.join(folder_path, doc_id)
        try:
            if not os.path.exists(file_path):
                print(f"Warning: Document file not found: {file_path}")
                texts.append("")
                continue
            with open(file_path, 'r', encoding='utf-8') as file:
                soup = BeautifulSoup(file, 'html.parser')
                text = soup.get_text(separator=' ', strip=True)
                texts.append(text)
        except Exception as e:
            print(f"Error retrieving document {doc_id}: {e}")
            texts.append("")
    return texts

def retrieve_rec_texts(
    document_indices, 
    folder_path='downloaded_articles/downloaded_articles', 
    metadata_path='recipes_metadata.xlsx'
):
    try:
        metadata_df = pd.read_excel(metadata_path)
        if "id" not in metadata_df.columns or "original_file_name" not in metadata_df.columns:
            raise ValueError("Metadata file must contain 'id' and 'original_file_name' columns.")
        metadata_df = metadata_df.sort_values(by="id").reset_index(drop=True)
        if metadata_df.index.max() < max(document_indices):
            raise ValueError("Some document indices exceed the range of metadata.")
        document_texts = []
        for idx in document_indices:
            if idx >= len(metadata_df):
                print(f"Warning: Index {idx} is out of range for metadata.")
                continue
            original_file_name = metadata_df.iloc[idx]["original_file_name"]
            if not original_file_name:
                print(f"Warning: No file name found for index {idx}")
                continue
            file_path = os.path.join(folder_path, original_file_name)
            if os.path.exists(file_path):
                with open(file_path, "r", encoding="utf-8") as f:
                    document_texts.append(f.read())
            else:
                print(f"Warning: File not found at {file_path}")
        return document_texts
    except Exception as e:
        print(f"Error in retrieve_rec_texts: {e}")
        return []

def retrieve_metadata(document_indices: List[int], metadata_path: str = 'recipes_metadata.xlsx') -> Dict[int, Dict[str, str]]:
    try:
        metadata_df = pd.read_excel(metadata_path)
        required_columns = {'id', 'original_file_name', 'url'}
        if not required_columns.issubset(metadata_df.columns):
            raise ValueError(f"Metadata file must contain columns: {required_columns}")
        metadata_df['id'] = metadata_df['id'].astype(int)  
        filtered_metadata = metadata_df[metadata_df['id'].isin(document_indices)]
        metadata_dict = {
            int(row['id']): {
                "original_file_name": row['original_file_name'],
                "url": row['url']
            }
            for _, row in filtered_metadata.iterrows()
        }
        return metadata_dict
    except Exception as e:
        print(f"Error retrieving metadata: {e}")
        return {}

def rerank_documents(query, document_ids, document_texts, cross_encoder_model):
    try:
        pairs = [(query, doc) for doc in document_texts]
        scores = cross_encoder_model.predict(pairs)
        scored_documents = list(zip(scores, document_ids, document_texts))
        scored_documents.sort(key=lambda x: x[0], reverse=True)
        print("Reranked results:")
        for idx, (score, doc_id, doc) in enumerate(scored_documents):
            print(f"Rank {idx + 1} (Score: {score:.4f}, Document ID: {doc_id})")
        return scored_documents
    except Exception as e:
        print(f"Error reranking documents: {e}")
        return []

def translate_ar_to_en(text):
    try:
        ar_to_en_tokenizer = models['ar_to_en_tokenizer'] = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
        ar_to_en_model= models['ar_to_en_model'] = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
        inputs = ar_to_en_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
        translated_ids = ar_to_en_model.generate(
            inputs.input_ids,
            max_length=512, 
            num_beams=4,     
            early_stopping=True
        )
        translated_text = ar_to_en_tokenizer.decode(translated_ids[0], skip_special_tokens=True)
        return translated_text
    except Exception as e:
        print(f"Error during Arabic to English translation: {e}")
        return None
              
def translate_en_to_ar(text):
    try:
        en_to_ar_tokenizer = models['en_to_ar_tokenizer'] = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ar")
        en_to_ar_model = models['en_to_ar_model'] = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-en-ar")  
        inputs = en_to_ar_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
        translated_ids = en_to_ar_model.generate(
            inputs.input_ids,
            max_length=512,  
            num_beams=4,     
            early_stopping=True
        )
        translated_text = en_to_ar_tokenizer.decode(translated_ids[0], skip_special_tokens=True)
        return translated_text
    except Exception as e:
        print(f"Error during English to Arabic translation: {e}")
        return None


@app.get("/")
async def root():
    return {"message": "Welcome to TeaRAG! Your Medical Assistant Powered by RAG"}

@app.get("/health")
async def health_check():
    """Health check endpoint"""
    status = {
        'status': 'healthy',
        'models_loaded': bool(models),
        'embeddings_loaded': bool(data.get('embeddings')),
        'documents_loaded': not data.get('df', pd.DataFrame()).empty
    }
    return status

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)