Open Models Almanac
Methodspost-trainingrollout & RL infrastructure

implementation detail · filed under post-training

Isolated resettable sandboxes for agent rollouts

Runs each rollout in an isolated sandbox that can be restored to a fixed initial state.

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.

sources
2
models
3
labs adopt it
2
strongest
used

Documented in

Evidence

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

How sources treat it
used 2

Kolibri-1

Aleph Alphaadopted1 source

To isolate it from the node and from other sandboxes, the instance runs in an unprivileged user namespace with its own mount, process and network namespaces.

usedsoftware implementationin Kolibri RL training runAleph Alpha
↑