src/utils/llm_utils.py

# src/utils/llm_utils.py
from fastapi import HTTPException
from typing import Tuple

from src.llms.openai_llm import OpenAILanguageModel
from src.llms.ollama_llm import OllamaLanguageModel
from src.llms.bert_llm import BERTLanguageModel
from src.llms.falcon_llm import FalconLanguageModel
from src.llms.llama_llm import LlamaLanguageModel
from src.embeddings.huggingface_embedding import HuggingFaceEmbedding
from src.vectorstores.chroma_vectorstore import ChromaVectorStore
from src.vectorstores.optimized_vectorstore import get_optimized_vector_store
from src.utils.logger import logger
from config.config import settings

def get_llm_instance(provider: str):
    """
    Get LLM instance based on provider
    
    Args:
        provider (str): Name of the LLM provider
        
    Returns:
        BaseLLM: Instance of the LLM
        
    Raises:
        ValueError: If provider is not supported
    """
    llm_map = {
        'openai': lambda: OpenAILanguageModel(api_key=settings.OPENAI_API_KEY),
        'ollama': lambda: OllamaLanguageModel(base_url=settings.OLLAMA_BASE_URL),
        'bert': lambda: BERTLanguageModel(),
        'falcon': lambda: FalconLanguageModel(),
        'llama': lambda: LlamaLanguageModel(),
    }
    
    if provider not in llm_map:
        raise ValueError(f"Unsupported LLM provider: {provider}")
    return llm_map[provider]()

async def get_vector_store() -> Tuple[ChromaVectorStore, HuggingFaceEmbedding]:
    """
    Get vector store and embedding model instances
    Uses optimized implementation while maintaining backward compatibility
    
    Returns:
        Tuple[ChromaVectorStore, HuggingFaceEmbedding]: 
            Vector store and embedding model instances
    """
    try:
        return await get_optimized_vector_store()
    except Exception as e:
        logger.error(f"Error getting optimized vector store: {str(e)}")
        # Fallback to original implementation if optimization fails
        logger.warning("Falling back to standard vector store implementation")
        embedding = HuggingFaceEmbedding(model_name=settings.EMBEDDING_MODEL)
        vector_store = ChromaVectorStore(
            embedding_function=embedding.embed_documents,
            persist_directory=settings.CHROMA_PATH
        )
        return vector_store, embedding