utils/models.py

import torch
import chromadb
from transformers import CLIPProcessor, CLIPModel, AutoModelForCausalLM, AutoTokenizer
from utils.config import config, get_logger

logger = get_logger("Models")
device = "cuda" if torch.cuda.is_available() else "cpu"
logger.info(f"Initializing models on: {device.upper()}")

# 1. Load CLIP
logger.info(f"Loading CLIP ({config.clip_model_id})...")
clip_processor = CLIPProcessor.from_pretrained(config.clip_model_id)
clip_model = CLIPModel.from_pretrained(config.clip_model_id).to(device)

# 2. Initialize ChromaDB
logger.info("Initializing ChromaDB...")
chroma_client = chromadb.Client()
try:
    chroma_client.delete_collection(config.collection_name)
except Exception:
    pass
collection = chroma_client.create_collection(name=config.collection_name)

# --- THE MONKEY PATCH ---
# Intercept the breaking change in Transformers v4.45+ for custom models
_orig_getattr = torch.nn.Module.__getattr__

def _patched_getattr(self, name):
    if name == "all_tied_weights_keys":
        return {}
    return _orig_getattr(self, name)

torch.nn.Module.__getattr__ = _patched_getattr
# ------------------------

# 3. Load VLM
logger.info(f"Loading VLM ({config.vlm_model_id})...")
vlm_model = AutoModelForCausalLM.from_pretrained(
    config.vlm_model_id, trust_remote_code=True, revision=config.vlm_revision
).to(device)
vlm_tokenizer = AutoTokenizer.from_pretrained(config.vlm_model_id, revision=config.vlm_revision)