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

build_greedy_acceptance_sampler_graph

build_greedy_acceptance_sampler_graph()

max.pipelines.sampling.build_greedy_acceptance_sampler_graph(device)

source

Builds a graph that implements strict greedy acceptance for MTP.

Draft tokens are accepted only when they match the argmax of the target logits at each position. Always produces a recovered token for every draft position and a bonus token from the final (+1) target position.

Parameters:

device (DeviceRef) – Device for the graph.

Returns:

A graph that takes draft tokens, target logits, and target logit offsets and outputs the first rejected index, target tokens for all draft positions, and a bonus token.

Return type:

Graph