diff --git a/train_sft.py b/train_sft.py
index 656eca5..56618bc 100644
--- a/train_sft.py
+++ b/train_sft.py
@@ -348,7 +348,7 @@ def main() -> None:
                             },
                             step=step,
                         )
-                    if _is_better_eval(
+                    if step > 0 and _is_better_eval(
                         printable["success_rate"],
                         printable["chrf"],
                         best_success,
@@ -362,6 +362,8 @@ def main() -> None:
                             f"chrf {best_chrf:.2f} -> {best_path}"
                         )
                     model.train()
+                    if device == "cuda":
+                        torch.cuda.empty_cache()
             if ckpt_now:
                 _save(last_path, dump())
                 print(f"   last -> {last_path}")