Open Models Almanac
Methodstraining objectivedistillation objective

specific method · filed under training objective

Skew reverse-KL distillation

Distillation uses a skew reverse-KL loss variant.

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

Also called Skew Reverse KL divergence, SkewReverse loss for distillation, SkewReverse loss.

sources
2
models
2
lab adopt it
1
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

Ultra’s coding capabilities, combined with the SkewReverse loss used during distillation, helped improve coding accuracy.

usedtraining objectivein Nemotron 3.5 Super VLNVIDIA
1 more from this model

with Ultra 3 Base as the teacher and Skew Reverse KL divergence as the distillation loss function

usedtraining objectivein Nemotron 3.5 SuperNVIDIA
↑