diff --git a/parso/python/tokenize.py b/parso/python/tokenize.py
index 63abff3..64610dd 100644
--- a/parso/python/tokenize.py
+++ b/parso/python/tokenize.py
@@ -52,7 +52,7 @@ class TokenCollection(NamedTuple):
 
 BOM_UTF8_STRING = BOM_UTF8.decode('utf-8')
 
-_token_collection_cache: Dict[tuple[int, int], TokenCollection] = {}
+_token_collection_cache: Dict[Tuple[int, int], TokenCollection] = {}
 
 
 def group(*choices, capture=False, **kwargs):
@@ -340,7 +340,7 @@ def _find_fstring_string(endpats, fstring_stack, line, lnum, pos):
 
 
 def tokenize(
-    code: str, *, version_info: PythonVersionInfo, start_pos: Tuple[int, int] = (1, 0)
+    code: str, *, version_info: Tuple[int, int], start_pos: Tuple[int, int] = (1, 0)
 ) -> Iterator[PythonToken]:
     """Generate tokens from a the source code (string)."""
     lines = split_lines(code, keepends=True)
@@ -363,7 +363,7 @@ def _print_tokens(func):
 def tokenize_lines(
     lines: Iterable[str],
     *,
-    version_info: PythonVersionInfo,
+    version_info: Tuple[int, int],
     indents: List[int] = None,
     start_pos: Tuple[int, int] = (1, 0),
     is_first_token=True,
diff --git a/parso/utils.py b/parso/utils.py
index 6ad1ea2..7ff342a 100644
--- a/parso/utils.py
+++ b/parso/utils.py
@@ -2,7 +2,7 @@ import re
 import sys
 from ast import literal_eval
 from functools import total_ordering
-from typing import NamedTuple, Sequence, Union
+from typing import NamedTuple, Union
 
 # The following is a list in Python that are line breaks in str.splitlines, but
 # not in Python. In Python only \r (Carriage Return, 0xD) and \n (Line Feed,
@@ -26,7 +26,7 @@ class Version(NamedTuple):
     micro: int
 
 
-def split_lines(string: str, keepends: bool = False) -> Sequence[str]:
+def split_lines(string: str, keepends: bool = False) -> "list[str]":
     r"""
     Intended for Python code. In contrast to Python's :py:meth:`str.splitlines`,
     looks at form feeds and other special characters as normal text. Just
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..6afd8e5
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,14 @@
+[tool.zuban]
+enable_error_code = ["ignore-without-code"]
+
+disallow_subclassing_any = true
+
+# Avoid creating future gotchas emerging from bad typing
+warn_redundant_casts = true
+warn_unused_ignores = true
+warn_unused_configs = true
+warn_unreachable = true
+
+strict_equality = true
+implicit_optional = true
+exclude = "^test/normalizer_issue_files"
diff --git a/test/test_load_grammar.py b/test/test_load_grammar.py
index 0c70436..4a03a0f 100644
--- a/test/test_load_grammar.py
+++ b/test/test_load_grammar.py
@@ -33,4 +33,4 @@ def test_invalid_grammar_version(string):
 
 def test_grammar_int_version():
     with pytest.raises(TypeError):
-        load_grammar(version=3.8)
+        load_grammar(version=3.8)  # type: ignore