Update app.py

app.py

@@ -3,7 +3,7 @@ from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 from typing import List, Optional
 import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer, LlamaTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, LlamaTokenizerFast
 import re
 import os
 
@@ -29,7 +29,7 @@ HF_TOKEN = os.getenv("HF_TOKEN")  # Load from Render environment variable
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-tokenizer = AutoTokenizer.from_pretrained(REPO_NAME, token=HF_TOKEN)
+tokenizer = LlamaTokenizerFast.from_pretrained(REPO_NAME, token=HF_TOKEN)
 #tokenizer = LlamaTokenizer.from_pretrained(REPO_NAME, use_auth_token=HF_TOKEN)
 
 model = AutoModelForCausalLM.from_pretrained(