Open Models Almanac
Methodspost-trainingreinforcement learning algorithm

specific method · filed under post-training

Asynchronous Group Relative Policy Optimization

GRPO training performed asynchronously, including on large batches.

The sources on this page name it without explaining how it works. The line above is the extraction's one-line gloss. See reinforcement learning algorithm for the methods filed beside it.

Also called asynchronous GRPO, Asynchronous GRPO training, asynchronous GRPO framework.

sources
3
models
4
labs adopt it
2
strongest
core

Documented in

Evidence

5 quotes from 3 sources, one card per model, each opening with its strongest.

How sources treat it
used 4core 1

We initialize from an existing multimodal RLVR checkpoint and train using asynchronous GRPO

usedunclearin Nemotron 3.5 vision reasoning teacherNVIDIA
2 more from these models

We initialize from the step-120 text-RLVR checkpoint and train using asynchronous GRPO

usedunclearin Nemotron 3.5 video reasoning and tracking teacherNVIDIA

continue training with the same asynchronous GRPO framework.

usedpost trainingin Nemotron 3.5 SuperNVIDIA
↑