Open Models Almanac
Methodsoptimizationtraining runtime

general family · filed under optimization

Checkpoint merging

Combines checkpoints into a merged model; the evidence does not specify a single merging algorithm.

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

Also called Offline Checkpoint Merging.

sources
4
models
3
labs adopt it
3
strongest
used

Documented in

Evidence

5 quotes from 4 sources, one card per model, each opening with its strongest.

How sources treat it
used 5

Kolibri-1

Aleph Alphaadopted1 source

Checkpoint merging largely recovers this difference, with the four merged checkpoints indistinguishable within the standard error of the general-capability composite.

usedunclearin KolibriAleph Alpha
1 more from this model

Checkpoint merging largely preserves general capability.

usedunclearin KolibriAleph Alpha

DeepSeek-V4.1-Flash

DeepSeekadopted1 source

we merge checkpoints from runs across different scaffolds or configurations, combining improvements acquired along different optimization paths

usedunclearin DeepSeek-V4.1-FlashDeepSeek
↑