services/model_service.py

from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaConfig
from sentence_transformers import SentenceTransformer
import torch
from functools import lru_cache
from config.config import settings
import logging

logger = logging.getLogger(__name__)

class ModelService:
    _instance = None

    def __new__(cls):
        if cls._instance is None:
            cls._instance = super().__new__(cls)
            cls._instance._initialized = False
        return cls._instance

    def __init__(self):
        if not self._initialized:
            self._initialized = True
            self._load_models()

    @lru_cache(maxsize=1)
    def _load_models(self):
        try:
            # Load tokenizer
            self.tokenizer = AutoTokenizer.from_pretrained(settings.MODEL_NAME)

            # Load model configuration without modifying rope_scaling
            config = LlamaConfig.from_pretrained(settings.MODEL_NAME)

            # Remove rope_scaling if present
            if hasattr(config, "rope_scaling"):
                logger.info("Removing rope_scaling from configuration...")
                config.rope_scaling = None

            # Load model with the updated configuration
            self.model = AutoModelForCausalLM.from_pretrained(
                settings.MODEL_NAME,
                torch_dtype=torch.float16 if settings.DEVICE == "cuda" else torch.float32,
                device_map="auto" if settings.DEVICE == "cuda" else None,
                config=config
            )

            # Load sentence embedder
            self.embedder = SentenceTransformer(settings.EMBEDDER_MODEL)

        except Exception as e:
            logger.error(f"Error loading models: {e}")
            raise

    def get_models(self):
        return self.tokenizer, self.model, self.embedder