Model techniques map
Techniquespost-trainingrollout & RL infrastructure

general family · filed under post-training

Asynchronous reinforcement learning

An asynchronous RL training approach that overlaps rollout generation and training; some implementations explicitly decouple training from inference across devices.

Also called asynchronous RL architecture, asynchronous training paradigm, fully asynchronous training, Asynchronous training.

sources
6
models
5
labs adopt it
3
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 6core 1

Documented in

Further reading

Picked by hand, not extracted: where to read more, not evidence for anything on this page.

Evidence

7 spans quoted from the sources, strongest treatment first.

we scale batch size and training throughput through fully asynchronous training, processing thousands of long-horizon rollouts and billions of tokens per step

coresoftware implementationin MiMo-V2.6 RL trainingXiaomi

These advances are supported by asynchronous training, mixed-task rollout infrastructure, training–inference consistency mechanisms

usedoptimizationin MiMo-V2.6

It utilized an asynchronous RL architecture that fully decouples training from inference across separate GPU devices

usedoptimizationin Nemotron 3 UltraNVIDIA

It utilized an asynchronous RL architecture that decouples training from inference and leverages MTP to accelerate rollout generation.

usedpost trainingin Nemotron 3.5 LightningNVIDIA

asynchronous RL architecture that fully decouples training from inference across separate GPU devices

usedoptimizationin Nemotron 3 UltraNVIDIA

we adopt a fully asynchronous training paradigm for Agentic RL

usedtraining objectivein GLM-5Z.ai

we employ an asynchronous RL architecture that decouples training from inference and leverage multi-token prediction to accelerate rollout generation

usedpost trainingin Nemotron 3 familyNVIDIA

Filed alongside

Other methods under post-training :: rollout & RL infrastructure.