For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Python function

greedy_acceptance_sampler

greedy_acceptance_sampler()

max.nn.sampling.greedy_acceptance_sampler(draft_tokens, target_logits, token_bitmasks=None)

source

Target-only rejection sampler for speculative decoding.

Accepts a draft token only when it matches the argmax of the target logits. Recovered tokens are the target argmax at every draft position; the bonus token is the argmax at the final (+1) position.

When token_bitmasks is provided, grammar constraints mask the target logits (fill -inf) before the argmax, so a grammar-invalid draft is always rejected and recovered and bonus tokens always satisfy structured-output constraints — same contract as the stochastic path.

Returns (first_rejected_idx, recovered_tokens, bonus_tokens)

Parameters:

• draft_tokens (TensorValue)
• target_logits (TensorValue)
• token_bitmasks (TensorValue | None)

Return type:

tuple[TensorValue, TensorValue, TensorValue]