New API stack migration guide#

This page explains how to convert your existing old API stack RLlib classes and code to RLlib’s new API stack.

What’s the new API stack?#

The new API stack rewrites the core RLlib APIs from scratch and reduces user-facing classes from more than a dozen critical ones to only a handful, without losing features. When designing these interfaces, the Ray team strictly applied the following principles:

Classes must be usable outside of RLlib.
Separation of concerns. Answer the question “what should get done, when, and by whom?” and give each class as few non-overlapping, well-defined tasks as possible.
Offer fine-grained modularity, full interoperability, and frictionless pluggability of classes.
Use widely accepted third-party standards and APIs wherever possible.

Applying the preceding principles, the Ray team reduced the must-know classes from eight on the old stack to only five on the new stack. The core new API stack classes are:

RLModule, which replaces ModelV2 and PolicyMap APIs.
Learner, which replaces RolloutWorker and some of Policy.
SingleAgentEpisode and MultiAgentEpisode, which replace ViewRequirement, SampleCollector, Episode, and EpisodeV2.
ConnectorV2, which replaces Connector and some of RolloutWorker and Policy.

The AlgorithmConfig and Algorithm APIs remain as-is. These classes are already established APIs on the old stack.

Note

RLlib supports a single deep learning framework, PyTorch, and drops TensorFlow support entirely. The Ray team continues to design RLlib to be framework-agnostic and might add support for other frameworks.

Check your AlgorithmConfig#

RLlib turns on the new API stack by default for all RLlib algorithms.

Note

To deactivate the new API stack and switch back to the old one, use the api_stack() method in your AlgorithmConfig object as follows:

config.api_stack(
    enable_rl_module_and_learner=False,
    enable_env_runner_and_connector_v2=False,
)

There are a few other differences between configuring an old API stack algorithm and its new stack counterpart. Go through the following sections and translate the respective settings. Remove settings that the new stack doesn’t support or need.

AlgorithmConfig.framework()#

RLlib supports a single deep learning framework, PyTorch.

The new API stack deprecates the following framework-related settings:

# Make sure you always set the framework to "torch"...
config.framework("torch")

# ... and drop all tf-specific settings.
config.framework(
    eager_tracing=True,
    eager_max_retraces=20,
    tf_session_args={},
    local_tf_session_args={},
)

AlgorithmConfig.resources()#

The Ray team deprecated the num_gpus and _fake_gpus settings. To place your RLModule on one or more GPUs on the Learner side, do the following:

# The following setting is equivalent to the old stack's `config.resources(num_gpus=2)`.
config.learners(
    num_learners=2,
    num_gpus_per_learner=1,
)

Hint

The num_learners setting determines how many remote Learner workers there are in your Algorithm’s LearnerGroup. If you set this parameter to 0, your LearnerGroup only contains a local Learner that runs on the main process and shares its compute resources, typically 1 CPU. For asynchronous algorithms such as IMPALA or APPO, always set this parameter greater than 0.

For an example of training with fractional GPUs, see the fractional GPUs example script. For fractional GPUs, always set num_learners to 0 or 1.

If GPUs aren’t available, but you want to learn with more than one Learner in a multi-CPU fashion, do the following:

config.learners(
    num_learners=2,  # or >2
    num_cpus_per_learner=1,  # <- default
    num_gpus_per_learner=0,  # <- default
)

The Ray team renamed the setting num_cpus_for_local_worker to num_cpus_for_main_process.

config.resources(num_cpus_for_main_process=0)  # default is 1

AlgorithmConfig.training()#

Train batch size#

Because of the new API stack’s Learner worker architecture, training might happen in a distributed fashion over n Learner workers, so RLlib provides the train batch size per individual Learner. Don’t use the train_batch_size setting anymore:

config.training(
    train_batch_size_per_learner=512,
)

You don’t need to change this setting, even when increasing the number of Learner, through config.learners(num_learners=...).

A good rule of thumb for scaling on the learner axis is to keep the train_batch_size_per_learner value constant as the number of Learners grows and to increase the learning rate as follows:

lr = [original_lr] * ([num_learners] ** 0.5)

Neural network configuration#

The old stack’s config.training(model=...) is no longer supported on the new API stack. Instead, use the new rl_module() method to configure RLlib’s default RLModule or specify and configure a custom RLModule.

See RLModules API, a general guide that also explains the use of the config.rl_module() method.

If you have an old stack ModelV2 and want to migrate the entire NN logic to the new stack, see ModelV2 to RLModule for migration instructions.

Learning rate and coefficient schedules#

If you’re using schedules for learning rate or other coefficients, such as the entropy_coeff setting in PPO, provide the scheduling information directly in the respective setting. Scheduling behavior no longer requires a specific, separate setting.

When defining a schedule, provide a list of 2-tuples. The first item is the global timestep, and the second item is the value the learning rate should reach at that timestep. The reported metrics list this timestep as num_env_steps_sampled_lifetime. Always start the first 2-tuple with timestep 0. RLlib linearly interpolates values between two provided timesteps.

For example, to create a learning rate schedule that starts with a value of 1e-5, then increases over 1M timesteps to 1e-4 and stays constant after that, do the following:

config.training(
    lr=[
        [0, 1e-5],  # <- initial value at timestep 0
        [1000000, 1e-4],  # <- final value at 1M timesteps
    ],
)

In the preceding example, the value after 500k timesteps is roughly 5e-5 from linear interpolation.

As another example, to create an entropy coefficient schedule that starts at 0.05, increases over 1M timesteps to 0.1, and then suddenly drops to 0 right after the 1Mth timestep, do the following:

config.training(
    entropy_coeff=[
        [0, 0.05],  # <- initial value at timestep 0
        [1000000, 0.1],  # <- value at 1M timesteps
        [1000001, 0.0],  # <- sudden drop to 0.0 right after 1M timesteps
    ]
)

If you need to configure more complex learning rate scheduling behavior or chain different schedulers into a pipeline, use the experimental _torch_lr_schedule_classes config property. See this example script. This example covers learning rate schedules only, not other coefficients.

AlgorithmConfig.learners()#

This method isn’t used on the old API stack because the old stack doesn’t use Learner workers.

Use this method to specify the following:

the number of Learner workers through .learners(num_learners=...).
the resources per learner. Use .learners(num_gpus_per_learner=1) for GPU training and .learners(num_gpus_per_learner=0) for CPU training.
the custom Learner class you want to use. See this custom loss function example script.
a config dict to set for your custom learner: .learners(learner_config_dict={...}). Every Learner can access the entire AlgorithmConfig object through self.config, but setting learner_config_dict is a convenient way to avoid creating an entirely new AlgorithmConfig subclass to support a few extra settings for your custom Learner class.

AlgorithmConfig.env_runners()#

# RolloutWorkers have been replaced by EnvRunners. EnvRunners are more efficient and offer
# a more separation-of-concerns design and cleaner code.
config.env_runners(
    num_env_runners=2,  # use this instead of `num_workers`
)

# The following `env_runners` settings are deprecated and should no longer be explicitly
# set on the new stack:
config.env_runners(
    create_env_on_local_worker=False,
    sample_collector=None,
    enable_connectors=True,
    remote_worker_envs=False,
    remote_env_batch_wait_ms=0,
    preprocessor_pref="deepmind",
    enable_tf1_exec_eagerly=False,
    sampler_perf_stats_ema_coef=None,
)

Hint

If you want to IDE-debug what’s happening inside your EnvRunners, set num_env_runners=0 and run your experiment locally instead of through Ray Tune. To do this with any of RLlib’s example or tuned_example scripts, set the command-line arguments --no-tune --num-env-runners=0.

If you use the observation_filter setting, perform the following translations:

# For `observation_filter="NoFilter"`, don't set anything in particular. This is the default.

# For `observation_filter="MeanStdFilter"`, do the following:
from ray.rllib.connectors.env_to_module import MeanStdFilter

config.env_runners(
    env_to_module_connector=lambda env: MeanStdFilter(multi_agent=False),  # <- or True
)

Hint

The main switch for whether to explore during sample collection has moved to the env_runners() method. For details, see the exploration configuration.

AlgorithmConfig.exploration()#

The main switch for whether to explore during sample collection has moved from the deprecated AlgorithmConfig.exploration() method to env_runners().

This setting determines which method your RLModule calls inside the EnvRunner. It calls _forward_exploration() when explore=True and _forward_inference() when explore=False.

config.env_runners(explore=True)  # <- or False

The Ray team deprecated the exploration_config setting. Instead, define the exact exploratory behavior, such as sampling an action from a distribution, inside the overridden _forward_exploration() method of your RLModule.

Custom callbacks#

If you’re using custom callbacks on the old API stack, you’re subclassing the DefaultCallbacks class, which the Ray team renamed to RLlibCallback. You can continue this approach with the new API stack and pass your custom subclass to your config as follows:

# config.callbacks(YourCallbacksClass)

However, if you’re overriding methods that trigger on the EnvRunner side, such as on_episode_start/stop/step/etc..., you might have to translate some call arguments.

The following is a one-to-one translation guide for these types of RLlibCallback methods:

from ray.rllib.callbacks.callbacks import RLlibCallback

class YourCallbacksClass(RLlibCallback):

    def on_episode_start(
        self,
        *,
        episode,
        env_runner,
        metrics_logger,
        env,
        env_index,
        rl_module,

        # Old API stack args; don't use or access these inside your method code.
        worker=None,
        base_env=None,
        policies=None,
        **kwargs,
    ):
        # The `SingleAgentEpisode` or `MultiAgentEpisode` that RLlib has just started.
        # See https://docs.ray.io/en/latest/rllib/single-agent-episode.html for more details:
        print(episode)

        # The `EnvRunner` class that collects the episode in question.
        # This class used to be a `RolloutWorker`. On the new stack, this class is either a
        # `SingleAgentEnvRunner` or a `MultiAgentEnvRunner` holding the gymnasium Env,
        # the RLModule, and the 2 connector pipelines, env-to-module and module-to-env.
        print(env_runner)

        # The MetricsLogger object on the EnvRunner (documentation is a WIP).
        print(metrics_logger.peek("episode_return_mean", default=0.0))

        # The gymnasium env that sample collection uses. Note that this env may be a
        # gymnasium.vector.VectorEnv.
        print(env)

        # The env index, in case of a vector env, that handles the `episode`.
        print(env_index)

        # The RL Module that this EnvRunner uses. Note that this module may be a "plain", single-agent
        # `RLModule`, or a `MultiRLModule` in the multi-agent case.
        print(rl_module)

# Change similarly:
# on_episode_created()
# on_episode_step()
# on_episode_end()

The following callback methods are no longer available on the new API stack:

on_sub_environment_created(): The new API stack uses Farama’s gymnasium vector Envs, which leave RLlib no way to call a callback when each individual env-index is created.
on_create_policy(): This method is no longer available on the new API stack because only RolloutWorker calls it.
on_postprocess_trajectory(): The new API stack no longer calls this method because ConnectorV2 pipelines handle trajectory processing entirely. The documentation for ConnectorV2 is under development.

ModelV2 to RLModule#

If you’re using a custom ModelV2 class and want to translate the entire NN architecture and possibly action distribution logic to the new API stack, see RL Modules in addition to this section.

Also, see these example scripts on how to write a custom CNN-containing RLModule and how to write a custom LSTM-containing RLModule.

There are various options for translating an existing, custom ModelV2 from the old API stack, to the new API stack’s RLModule:

Move your ModelV2 code to a new, custom RLModule class. See RL Modules for details.
Use an Algorithm checkpoint or a Policy checkpoint from an old API stack training run with the new stack RLModule convenience wrapper.
Use an existing AlgorithmConfig object from an old API stack training run with the new stack RLModule convenience wrapper.

In more complex scenarios, you might have implemented custom policies to modify how models and distributions are constructed.

Translating Policy.compute_actions_from_input_dict#

This old API stack method, as well as compute_actions and compute_single_action, directly translate to _forward_inference() and _forward_exploration(). The RLModule guide explains how to implement this method.

Translating Policy.action_distribution_fn#

To translate action_distribution_fn, write the following custom RLModule code:

Same action dist. class

from ray.rllib.models.torch.torch_distributions import YOUR_DIST_CLASS


class MyRLModule(TorchRLModule):
    def setup(self):
        ...
        # Set the following attribute at the end of your custom `setup()`.
        self.action_dist_cls = YOUR_DIST_CLASS

Different action dist. classes

from ray.rllib.models.torch.torch_distributions import (
    YOUR_INFERENCE_DIST_CLASS,
    YOUR_EXPLORATION_DIST_CLASS,
    YOUR_TRAIN_DIST_CLASS,
)

        def get_inference_action_dist_cls(self):
            return YOUR_INFERENCE_DIST_CLASS

        def get_exploration_action_dist_cls(self):
            return YOUR_EXPLORATION_DIST_CLASS

        def get_train_action_dist_cls(self):
            return YOUR_TRAIN_DIST_CLASS

Translating Policy.action_sampler_fn#

To translate action_sampler_fn, write the following custom RLModule code:

from ray.rllib.models.torch.torch_distributions import YOUR_DIST_CLASS


class MyRLModule(TorchRLModule):

    def _forward_exploration(self, batch):
        computation_results = ...
        my_dist = YOUR_DIST_CLASS(computation_results)
        actions = my_dist.sample()
        return {Columns.ACTIONS: actions}

    # Maybe for inference, you would like to sample from the deterministic version
    # of your distribution:
    def _forward_inference(self, batch):
        computation_results = ...
        my_dist = YOUR_DIST_CLASS(computation_results)
        greedy_actions = my_dist.to_deterministic().sample()
        return {Columns.ACTIONS: greedy_actions}

Policy.compute_log_likelihoods#

Implement your custom RLModule’s _forward_train() method and return the Columns.ACTION_LOGP key together with the corresponding action log probabilities to pass this information to your loss functions, which your code calls after forward_train(). The loss logic can then access Columns.ACTION_LOGP.

Custom loss functions and policies#

If you’re using one or more custom loss functions or custom PyTorch optimizers to train your models, move the logic into the new API stack’s Learner class instead of customizing inside the old stack’s Policy class.

See Learner for details on how to write a custom Learner.

The following example scripts show how to write:

The new API stack doesn’t support the Policy class. On the old stack, this class holds a neural network, a connector, and one or more optimizers and losses. On the new stack, the RLModule replaces the neural network, the ConnectorV2 replaces the connector, and the Learner class replaces the optimizers and losses.

The RLModule API is more flexible than the old stack’s Policy API and provides a cleaner separation of concerns. Action inference runs on the EnvRunners, and updating runs on the Learner workers. It also scales better, supporting multi-GPU training on any Ray cluster and multi-node, multi-GPU training on the Anyscale platform.

Custom connectors#

If you’re using custom connectors from the old API stack, move your logic into the new ConnectorV2 API. Translate your agent connectors into env-to-module ConnectorV2 pieces and your action connectors into module-to-env ConnectorV2 pieces.

The ConnectorV2 documentation is under development.

The following examples show how to write ConnectorV2 pieces for the different pipelines: