Pre-commit

Author: RedTachyon

recipes/dev/grpo_full_finetune_distributed.py

@@ -990,7 +990,10 @@ def train(self) -> None:
                     self._is_rank_zero
                     and curr_epoch == 0
                     and self.profiler_profile_memory
-                    and idx == self.profiler_wait_steps + self.profiler_warmup_steps + self.profiler_active_steps
+                    and idx
+                    == self.profiler_wait_steps
+                    + self.profiler_warmup_steps
+                    + self.profiler_active_steps
                     and self._device.type == "cuda"
                 ):
                     torch.cuda.memory._record_memory_history(enabled=None)