src/cag/generation_model.py

import requests
import numpy as np
import os
import time
import warnings
from dotenv import load_dotenv
from src.cag.cache_manager import CacheManager
from src.cag.embedding_utils import EmbeddingUtils

# Suppress PyTorch Warnings
warnings.filterwarnings("ignore", message="Tried to instantiate class '__path__._path'")

load_dotenv()

class LLMIntegration:
    def __init__(self,cache_size=100, similarity_threshold=0.8):
        """Initialize the LLM Integration with API Key, Cache, and Embedding Utilities."""
        self.cache_manager = CacheManager(max_cache_size=cache_size)
        self.embedding_utils = EmbeddingUtils()
        self.similarity_threshold = similarity_threshold

    def generate_response(self, query,response):
        """Generate a response with cache checking and similarity matching."""
        query_key = self.cache_manager.normalize_key(query)

        # Check for cache match
        cached_response = self.cache_manager.get_from_cache(query_key)
        if cached_response:
            return f"Cache Hit! {cached_response}"

        # Generate query embedding
        query_embedding = self.embedding_utils.generate_embedding(query)

        # Check for approximate match
        best_match_key = self._find_best_match(query_embedding)
        if best_match_key:
            cached_response = self.cache_manager.get_from_cache(best_match_key)
            return f"Cache Hit! {cached_response}"

        # If no cache match, query the API
        response = response
        
        # ✅ Only cache successful responses
        if response :
            self.cache_manager.add_to_cache(query_key, response, embedding=query_embedding)
            return f"Cache Miss! {response}"
        else:
            return "**Error: Could not generate a response.**"

    def _find_best_match(self, query_embedding):
        """Find the best match in the cache using similarity checking."""
        best_match_key = None
        highest_similarity = 0

        for key in self.cache_manager.cache:
            cached_embedding = self.cache_manager.get_embedding(key)
            if cached_embedding is not None:
                similarity = self.embedding_utils.calculate_similarity(query_embedding, cached_embedding)
                if similarity > highest_similarity and similarity >= self.similarity_threshold:
                    best_match_key = key
                    highest_similarity = similarity
        return best_match_key