From 2b41f092688b12a01cd2d42f449452268afd198a Mon Sep 17 00:00:00 2001
From: Krrish Dholakia <krrishdholakia@gmail.com>
Date: Tue, 14 May 2024 16:38:45 -0700
Subject: [PATCH] fix(utils.py): add lru-cache logic to _select_tokenizer

speed up tokenizer load times
---
 litellm/utils.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/litellm/utils.py b/litellm/utils.py
index f1db8ce6a55..0220fecd733 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -12,10 +12,9 @@ import dotenv, json, traceback, threading, base64, ast
 import subprocess, os
 from os.path import abspath, join, dirname
 import litellm, openai
-
 import itertools
 import random, uuid, requests  # type: ignore
-from functools import wraps
+from functools import wraps, lru_cache
 import datetime, time
 import tiktoken
 import uuid
@@ -3881,6 +3880,7 @@ def get_replicate_completion_pricing(completion_response=None, total_time=0.0):
     return a100_80gb_price_per_second_public * total_time / 1000
 
 
+@lru_cache(maxsize=128)
 def _select_tokenizer(model: str):
     from importlib import resources