optionally define whether to use_fast tokenizer

tests/test_tokenizers.py

@@ -0,0 +1,31 @@
+"""
+Test cases for the tokenizer loading
+"""
+import unittest
+
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.models import load_tokenizer
+
+
+class TestTokenizers(unittest.TestCase):
+    """
+    test class for the load_tokenizer fn
+    """
+
+    def test_default_use_fast(self):
+        cfg = DictDefault({})
+        tokenizer = load_tokenizer("huggyllama/llama-7b", None, cfg)
+        assert "Fast" in tokenizer.__class__.__name__
+
+    def test_dont_use_fast(self):
+        cfg = DictDefault(
+            {
+                "tokenizer_use_fast": False,
+            }
+        )
+        tokenizer = load_tokenizer("huggyllama/llama-7b", None, cfg)
+        assert "Fast" not in tokenizer.__class__.__name__
+
+
+if __name__ == "__main__":
+    unittest.main()