From aefd4d74fa5fb3df85cad4a2146551f9a52b5d01 Mon Sep 17 00:00:00 2001
From: Wing Lian <wing.lian@gmail.com>
Date: Mon, 21 Aug 2023 17:22:06 -0400
Subject: [PATCH] better handling when no subjects

---
 src/axolotl/utils/callbacks.py | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/src/axolotl/utils/callbacks.py b/src/axolotl/utils/callbacks.py
index 2f632a294..9ea949a7c 100644
--- a/src/axolotl/utils/callbacks.py
+++ b/src/axolotl/utils/callbacks.py
@@ -129,9 +129,11 @@ def bench_eval_callback_factory(trainer, tokenizer):
         def transform_subject(example):
             # Split on ':' and trim whitespace
             parts = example["subject"].split(":")
-            first_part = parts[0].strip().lower()  # Lowercase the first part
+            first_part = (
+                parts[0].strip().lower().replace("-", "_")
+            )  # Lowercase the first part
             second_part = (
-                parts[1].strip().replace("-", "_")
+                parts[1].strip().replace("-", "_") if len(parts) > 1 else "all"
             )  # Replace hyphens with underscores
 
             # Return the transformed values