correct typo (#11393)

examples/flax/language-modeling/run_mlm_flax.py

@@ -590,7 +590,7 @@ if __name__ == "__main__":
     # Create learning rate scheduler
     # warmup_steps = 0 causes the Flax optimizer to return NaNs; warmup_steps = 1 is functionally equivalent.
     lr_scheduler_fn = create_learning_rate_scheduler(
-        base_learning_rate=training_args.learning_rate, warmup_steps=min(training_args.warmup_steps, 1)
+        base_learning_rate=training_args.learning_rate, warmup_steps=max(training_args.warmup_steps, 1)
     )
 
     # Create parallel version of the training and evaluation steps