Merge pull request #4165 from AI-Hypercomputer:bvandermoon-fix

PiperOrigin-RevId: 932784151

Author: Google-ML-Automation

src/maxtext/configs/post_train/rl.yml

@@ -23,8 +23,7 @@ sampler_devices_fraction: 0.5
 chips_per_vm: 4  # depends on hardware, for v5p this is 4
 num_trainer_slices: -1
 num_samplers_slices: -1
-# Only specify rollout_data_parallelism when you would like to use more than one model
-# replicas in rollout. If not specified, rollout_tensor_parallelism will be auto-determined.
+# If rollout_data_parallelism is not specified, it will be auto-determined.
 rollout_data_parallelism: -1
 rollout_tensor_parallelism: 1
 rollout_expert_parallelism: 1