Implement in subclasses of [`PreTrainedModel`] for custom behavior to prepare inputs in the generate method.

This function extracts the model-specific `inputs` for generation.

Eliminate ("mask out") chosen categorical outcomes by setting their probability to 0. :param masks: An optional boolean ndarray of c

Returns list of lines from the text file. Performs lowercasing and white-space tokenization on each line before returning. Args: t

Resets the environment and starts a new episode

Takes a step with the given action and returns (next state, reward, done, info)

Implement in subclasses of [`PreTrainedModel`] for custom behavior to adjust the logits in the generate method.

Get the current policy distribution given the observations. :param obs: :param action_masks: :return: the action dis

Retrieve action distribution given the latent codes. :param latent_pi: Latent code for the actor :return: Action distributio

Evaluate actions according to the current policy, given the observations. :param obs: :param actions: :retur

Adapts the KL coeff

Updates the observation using the given action

Write predictions to file.

Write single and multiple references to file.

Add adjusted column offsets to take into account multi-column cells.

Heuristic to find column headers.

Heuristic to find row headers.

This class returns a [`LogitsProcessorList`] list object that contains all relevant [`LogitsWarper`] instances used for multinomial s

Returns count of all ngrams of given order in sequence.

Yields tables from the table file. Tables are parsed into a list of tuples with tokenized entries. Args: table_file: String filename.

called by constructor and __iadd__ to avoid creating new instances.

Gets the observation dimension

Returns the first device of the policy. Used in the case of model parallel

Inputs: - a list of commongens to score, e.g.,: - device: which torch device to load model on, e.g., "cuda:3" Outputs: - P(