fix ppov2_trainer tensorboard log bugs (#1836)

huggingface · Jul 16, 2024 · 052a8e1 · 052a8e1
1 parent a2adfb8
commit 052a8e1
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/trl/trainer/ppov2_trainer.py b/trl/trainer/ppov2_trainer.py
@@ -498,6 +498,7 @@ def repeat_generator():
                 metrics["lr"] = self.lr_scheduler.get_last_lr()[0]
                 metrics["episode"] = global_step
                 self.state.epoch = global_step / self.train_dataset_len  # used by self.log
+                self.state.global_step += 1
                 self.log(metrics)
             del kl, mean_kl, mean_entropy, mean_non_score_reward, scores, metrics, non_score_reward
             torch.cuda.empty_cache()