reference/generated/torchrl.data.vla.ActionTokenizerBase

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

ActionTokenizerBase

class torchrl.data.vla.ActionTokenizerBase(*args: Any, **kwargs: Any)

Base class for action tokenizers.

An action tokenizer maps continuous actions to discrete token ids and back, so that autoregressive (RT-2 / OpenVLA-style) VLA policies can emit actions through a language-model head and be trained with token cross-entropy.

A tokenizer operates element-wise over the trailing action dimension, so it works unchanged on per-step actions [*B, action_dim] and on action chunks [*B, T, chunk, action_dim].

Subclasses implement encode(), decode() and the vocab_size property.

decode(tokens: Tensor) → Tensor

Map token ids back to continuous actions [..., action_dim].

encode(actions: Tensor) → Tensor

Map continuous actions [..., action_dim] to token ids (long).

property vocab_size: int

Number of distinct token ids the tokenizer can emit per position.