Model techniques map
Techniquestraining objectiveauxiliary loss

specific method · filed under training objective

Domain-specific KL regularization strength

Uses different strengths of KL regularization for different domains.

Also called Domain-adaptive KL regularization strength, varying strengths of KL regularization.

source
1
model
1
lab adopt it
1
strongest
used

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 2

Documented in

Evidence

2 spans quoted from the sources, strongest treatment first.

In practice, we find that different domains benefit from varying strengths of KL regularization.

usedtraining objectivein DeepSeek-V3.2DeepSeek

different domains benefit from varying strengths of KL regularization.

usedoptimizationin DeepSeek-V3.2DeepSeek

Filed alongside

Other methods under training objective :: auxiliary loss.