diff --git a/tokenizer/rwkv_tokenizer.py b/tokenizer/rwkv_tokenizer.py
index a445950b..3ea0b489 100644
--- a/tokenizer/rwkv_tokenizer.py
+++ b/tokenizer/rwkv_tokenizer.py
@@ -217,14 +217,15 @@ def printTokens(self, tokens):
 # Tokenizer #4 (fast) https://github.com/LoganDark
 ########################################################################################################
 
+from typing import Generator
 from ast import literal_eval
 
 class FastTokenizer:
     __slots__ = ('tok2val', 'tok2len', 'root')
 
     def __init__(self, file_name):
-        self.tok2val = [b''] * 65536
-        self.tok2len = [0] * 65536
+        self.tok2val = {}
+        self.tok2len = {}
         self.root = {}
 
         with open(file_name, 'rt', encoding = 'utf-8') as file:
@@ -255,7 +256,7 @@ def next_token(self, src: bytes) -> int:
                 break
         return last_token
 
-    def encode_bytes(self, src: bytes) -> list[int]:
+    def encode_bytes(self, src: bytes) -> Generator[int, None, None]:
         start, stop = 0, len(src)
         while start < stop:
             last_token, last = None, self.root
@@ -272,9 +273,9 @@ def encode_bytes(self, src: bytes) -> list[int]:
             else: break
 
     def decode_bytes(self, tokens: list[int]) -> bytes:
-        return b''.join(map(self.tok2val.__getitem__, tokens))
+        return b''.join(map(self.tok2val.get, tokens))
 
-    def encode(self, src: str) -> list[int]:
+    def encode(self, src: str) -> Generator[int, None, None]:
         return self.encode_bytes(src.encode('utf-8'))
 
     def decode(self, tokens: list[int]) -> str: