fix eval

2025-09-04 10:34:41 +00:00 · 2024-07-11 03:35:03 +00:00
parent 8a9721bafe
commit e7a8634636
10 changed files with 171 additions and 41 deletions
--- a/applications/ColossalChat/coati/trainer/sft.py
+++ b/applications/ColossalChat/coati/trainer/sft.py
@@ -167,6 +167,7 @@ class SFTTrainer(SLTrainer):
            for tag in ["loss"]:
                msg = msg + f"{tag}: {self.accumulative_meter.get(tag)}\n"
            self.coordinator.print_on_master(msg)
+            os.makedirs(self.save_dir, exist_ok=True)
            with open(os.path.join(self.save_dir, f"eval_result_epoch{epoch}.txt"), "w") as f:
                f.write(msg)
            step_bar.close()