We use a temperature-scaled forward-KL loss against the backbone’s logits.
1 more from this model
We use a temperature-scaled forward-KL loss against the backbone’s logits.
specific method · filed under training objective
A forward-KL loss with temperature is computed against backbone logits.
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 temperature-scaled forward-KL loss.
2 quotes from 2 sources, one card per model, each opening with its strongest.
We use a temperature-scaled forward-KL loss against the backbone’s logits.
We use a temperature-scaled forward-KL loss against the backbone’s logits.