🧬 Fix typo in grpo_trainer.py (#2988)

huggingface · Feb 28, 2025 · 1a303cc · 1a303cc
1 parent ac327d5
commit 1a303cc
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/trl/trainer/grpo_trainer.py b/trl/trainer/grpo_trainer.py
@@ -574,7 +574,7 @@ def _get_train_sampler(self) -> Sampler:
         #    distributed to different GPUs, allowing rewards to be computed and normalized correctly within each prompt
         #    group. Using the same seed across processes ensures consistent prompt assignment, preventing discrepancies
         #    in group formation.
-        # 2. repeats the batch multiple times to allow reusing generaations across multiple updates. Refer to
+        # 2. repeats the batch multiple times to allow reusing generations across multiple updates. Refer to
         #    _prepare_inputs to see how the generations are stored and reused.
 
         # In the following figure, the values are the prompt indices. The first row shows the first sampled batch, the