diff --git a/model/supervised_finetuning/tests/test_utils.py b/model/supervised_finetuning/tests/test_utils.py
index ad40e534..c4982024 100644
--- a/model/supervised_finetuning/tests/test_utils.py
+++ b/model/supervised_finetuning/tests/test_utils.py
@@ -1,9 +1,28 @@
 from argparse import Namespace
 
-from utils import get_tokenizer
+import pytest
+from utils import TOKENIZER_CONFIGS, get_tokenizer, match_tokenizer_name
 
 
 def test_tokenizer():
     get_tokenizer(Namespace(model_name="Salesforce/codegen-2B-multi", cache_dir=".cache"))
     get_tokenizer(Namespace(model_name="facebook/galactica-1.3b", cache_dir=".cache"))
     get_tokenizer(Namespace(model_name="", cache_dir=".cache"))
+
+
+def test_tokenizer_successful_match():
+    for config_name, config in TOKENIZER_CONFIGS.items():
+        found_config = match_tokenizer_name(config_name)
+        assert found_config == config
+
+
+def test_tokenizer_partial_match():
+    for config_name in ["facebook/galactica-1.3b", "togethercomputer/GPT-JT-6B-v1", "Salesforce/codegen-2B-multi"]:
+        found_config = match_tokenizer_name(config_name)
+        assert found_config
+
+
+def test_tokenizer_failed_match():
+    for fake_config_name in ["not-a-model", "fake"]:
+        with pytest.raises(ValueError):
+            match_tokenizer_name(fake_config_name)
diff --git a/model/supervised_finetuning/utils.py b/model/supervised_finetuning/utils.py
index 380bca8e..1f77708f 100644
--- a/model/supervised_finetuning/utils.py
+++ b/model/supervised_finetuning/utils.py
@@ -106,7 +106,10 @@ TOKENIZER_CONFIGS = {
 
 
 def match_tokenizer_name(model_name: str) -> TokenizerConfig:
-    """Match a partial model name to a tokenizer configuration"""
+    """
+    Match a partial model name to a tokenizer configuration
+    i.e. model_name `Salesforce/codegen-2B-multi` has config name `codegen`
+    """
     tokenizer_config_matches = [config for name, config in TOKENIZER_CONFIGS.items() if name in model_name]
     if not tokenizer_config_matches:
         raise ValueError(f"Cannot find any tokeniser configuration to match {model_name=}")