Open Models Almanac
Methodspost-trainingrollout & RL infrastructure

specific method · filed under post-training

Co-located rollout and training

Runs rollout and training on the same physical devices, time-sharing their execution in the explicitly described implementation.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See rollout & RL infrastructure for the methods filed beside it.

Also called co-located RL training.

sources
2
models
2
labs adopt it
2
strongest
used

Documented in

Moonshot AI

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
used 2

Kimi K3

Moonshot AIadopted1 source

We adopt co-located RL training [58] to keep each 1M-context Kimi K3 RL experiment within a few hundred GPUs

usedpost trainingin Kimi K3Moonshot AI
↑