RolloutSkip Function Usage Documentation

Last updated: 2026-03-25

Applicable Scenarios

The RolloutSkip utility accelerates RL training by caching and reusing pre-generated rollout data, avoiding redundant sequence generation during debugging, replay, or fixed-experiment runs.

It is suitable for:

1. Re-running experiments with the same configuration
2. Speeding up training by skipping repeated generation
3. Reproducing rollout results in debugging

API and Usage Example

Trainer Adaptation

RolloutSkip is already supported in ``RayDAPOTrainer`` and ``RayPPOTrainer``.

Example integration:

.. code-block:: python

    from verl.utils.rollout_skip import RolloutSkip

    # Inside trainer.fit()
    rollout_skip = RolloutSkip(self.config, self.async_rollout_manager)
    rollout_skip.wrap_generate_sequences()


Basic Configuration

Add these parameters to enable RolloutSkip:

.. code-block:: bash

actor_rollout_ref.rollout.skip.enable=True
actor_rollout_ref.rollout.skip.dump_dir=/path/to/rollout_dump
actor_rollout_ref.rollout.skip.max_dump_step=10

Configuration Parameters

• skip.enable: Enable or disable RolloutSkip.
• skip.dump_dir: Root directory to save cached rollout data.
• skip.max_dump_step: Maximum number of steps to cache.

Cached Directory Structure

The directory structure is automatically generated to isolate different experiments:

.. code-block:: text

{dump_dir}/{exp_name}_{project_name}/
└── GBS{gbs}_N{n}_in{prompt_len}_out{response_len}/
    ├── train_step__gen_step.txt
    ├── genstep_000001/
    │   ├── new_batch.dp
    │   ├── gen_batch.dp
    │   └── meta.json
    └── genstep_000002/

Each genstep_* folder contains:

• new_batch.dp: Input prompt batch
• gen_batch.dp: Generated response batch
• meta.json: Step metadata