Add lazy LLM loading to fix ZeroGPU startup

model.py

@@ -2,7 +2,7 @@ import torch
 import numpy as np
 import pandas as pd
 from sentence_transformers import SentenceTransformer, util
-from transformers import AutoTokenizer, AutoModelForCausalLM
+
 import os
 
 # Load saved embeddings
@@ -15,17 +15,19 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
 print(f"[INFO] Using device: {device}")
 
 # Load embedding model
-embedding_model = SentenceTransformer("all-mpnet-base-v2", device=device)
+embedding_model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2", device=device)
 
 # Lazy-load the LLM model
 llm_model = None
 tokenizer = None
-HF_TOKEN = os.getenv("HF_TOKEN")
-model_id = "google/gemma-2-2b-it"
+
 
 def load_llm():
     global llm_model, tokenizer
     if llm_model is None or tokenizer is None:
+        from transformers import AutoTokenizer, AutoModelForCausalLM
+        HF_TOKEN = os.getenv("HF_TOKEN")
+        model_id = "google/gemma-2-2b-it"
         print("[INFO] Loading LLM model:", model_id)
         tokenizer = AutoTokenizer.from_pretrained(model_id, token=HF_TOKEN)
         llm_model = AutoModelForCausalLM.from_pretrained(