Limits how many steps to perform in the NAR-len (demask) pass.

Limits how long an utterance can be.

Repeats/trims the input prompt down to X seconds (0 to disable).

How many prior lines to serve as the context/prefix (0 to disable).

Adjusts the probabilities in the AR/NAR-len. (0 to greedy* sample)

Adjusts the probabilities in the NAR. (0 to greedy sample)

Classifier Free Guidance scale (AR needs 1, NAR-len needs 3).

Factor when rescaling for Classifier Free Guidance (0 to disable).

Filter out logits lower than this value.

Limits the samples that are outside the top P% of probabilities.

Limits the samples to the top K of probabilities.

Performs top-nσ logits processing.

Incurs a penalty to tokens based on how often they appear in a sequence.

Modifies the reptition penalty based on how far back in time the token appeared in the sequence.

(AR only) Modifies the probability of a stop token based on the current length of the sequence.

Modifies the randomness from the samples in the AR. (0 to greedy sample)

Limits the samples that are outside the top P% of probabilities.

Limits the samples to the top K of probabilities.

Number of branches to search through for beam search sampling.

Incurs a penalty to tokens based on how often they appear in a sequence.

Modifies the reptition penalty based on how far back in time the token appeared in the sequence.

(AR only) Modifies the probability of a stop token based on the current length of the sequence.