Open Models Almanac
Methodspost-trainingsupervised fine-tuning

implementation detail · filed under post-training

Assistant-only loss

Restricts supervised fine-tuning loss to assistant-generated content.

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

Also called assistant_only_loss=True, cross-entropy loss on the assistant turns.

sources
2
models
2
labs adopt it
2
strongest
used

Documented in

Aleph Alpha
Mistral AI

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

SFT of Kolibri proceeds after the long-context training stage with cross-entropy loss on the assistant turns

usedunclearin KolibriAleph Alpha
↑