A DataProto is a data structure that aims to provide a standard protocol for data exchange between functions. It contains a batch (TensorDict

Exception raised for errors in the parsing of an action.

Exception raised for errors in the parsing of an action.

A wrapper class for Ray actors with initialization arguments. This class extends ClassWithInitArgs to provide additional functionality for co

A unified tracking interface for logging experiment data to multiple backends. This class provides a centralized way to log experiment metrics, p

Represents an episode of an agent interacting with the environment. Attributes: done: Whether the agent indicated the task is complete. ste

Represents an episode of an agent interacting with the environment. Attributes: done: Whether the agent indicated the task is complete. ste

Sharding manager to support data resharding when using FSDP + Ulysses

Used to count mfu during training loop Example: flops_counter = FlopsCounter(config) flops_achieved, flops_promised = flops_

A schema property.

A schema property.

Define a resource pool specification. Resource pool will be initialized first.

The data model for async rollout.

Class for representing a bounding box.

Class for representing a bounding box.

Model loader that will set model weights to random values.

Model loader that will set model weights to random values.

Manage FSDP checkpointing in SPMD training. - Saves/loads per-rank sharded model & optimizer states - Persists full lr_scheduler and RNG

An LLM for generating texts from given prompts and sampling parameters. This class includes a tokenizer, a language model (possibly distributed

Mock multimodal LLM wrapper for testing.

Mock multimodal LLM wrapper for testing.

DataProtoFuture aims to eliminate actual data fetching on driver. By doing so, the driver doesn't have to wait for data so that asynchronous

Keeping track of all information across the agents.

Keeping track of all information across the agents.

Keeping track of all information across the agents.

A checkpoint manager that saves and loads - model - optimizer - lr_scheduler - extra_states in a SPMD way. We save -

Model loader that can load the model weights from partitioned megatron model.

Model loader that can load the model weights from partitioned megatron model.

Mock LLM wrapper for testing.

Mock LLM wrapper for testing.

The parsed schema of a tool in OpenAI format.

The tool call in OpenAI format.

Represents a UI element.

Represents a UI element.

Task to create a playlist in VLC.

Task to create a playlist in VLC.

Adaptive KL controller described in the paper: https://arxiv.org/pdf/1909.08593.pdf

Wrapper around ADB commands for device interaction.

Client for interacting with the Android environment server.

Client for interacting with the Android environment server.

Controller for an Android instance that adds accessibility tree data. The Accessibility Tree in Android is a tree-based structure, originally for

Controller for an Android instance that adds accessibility tree data. The Accessibility Tree in Android is a tree-based structure, originally for

Compared to synchronize, this uses more memory because of the buffer but achieves better performance due to the overlapping. D2h and h2d copying a