Returns the gradients of the passed sequence of parameters into a flat gradient. Order of parameters is preserved. :param output: functi

Checks whether gym env exposes a method returning invalid action masks :param env: the Gym environment to get masks from :return: A nump

Get tables in PARENT format for each json example.

Get references from json example.

r""" Generates sequences of token ids for models with a language modeling head using **greedy decoding** and can be used for text-deco

r""" Generates sequences of token ids for models with a language modeling head using **diverse beam search decoding** and can be used

Checks whether gym env exposes a method returning invalid action masks :param env: the Gym environment to check :return: True if the met

Return an instance of Distribution for the correct type of action space :param action_space: the input action space :return: the appropr

Metric for comparing predictions to references given tables. Args: predictions: An iterator over tokenized predictions. Each prediction is

Optional args for main function, mostly just to test.

Get the estimated values according to the current policy given the observations. :param obs: :return: the estimated values.

Create the layer that represents the distribution: it will be the logits of the Categorical distribution. You can then get pr

Renders the current state of the environment

Sets the observation featurizer (can also change during run time)

Recursively splits the given object

Update policy using the currently gathered rollout buffer.

Unpacks vectorized dict observations into separate dict observations

Callable for reward functions for text generation Args: current_observation (Observation): previous observation (s)

An abstract class for batched reward functions for text generation

[summary] Args: observation (Observation): current observation at t action (str): current action at t tar

add an instance (e.g., from another sentence).

Creates a priority sampler Args: max_size (int): maximum size of the queue priority_scale (float): 0.0 is a