Update hivemind/optim/averaged.py

Co-authored-by: Max Ryabinin <mryabinin0@gmail.com>

hivemind/optim/averaged.py

@@ -32,7 +32,7 @@ class DecentralizedOptimizer(DecentralizedOptimizerBase):
     :param average_gradients: whether to average gradients
     :param max_allowed_epoch_difference: if max_epoch has difference with local_epoch more than that, we download state
       from other peer.
-    :param total_steps_in_epoch: how many total steps must be to increase local_epoch by one
+    :param total_steps_in_epoch: the number of optimizer steps for a single training epoch
     :param average_opt_statistics: if specified, average optimizer states with corresponding names in state_dict
     :param scheduler_cls: a function which takes an optimizer and returns a learning rate scheduler
     :param averaging_steps_period: performs averaging after this many optimizer steps