implementation detail · filed under training objective
Full-vocabulary logit distillation
Uses the teacher’s full-vocabulary logits as the distributional target for the student.
- sources
- 2
- models
- 2
- 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.
we adopt full-vocabulary logit distillation in our OPD.
usedpost trainingin DeepSeek-V4DeepSeek
V4 uses full-vocabulary logit distillation instead.
usedpost trainingin DeepSeek-V4DeepSeek
Filed alongside
Other methods under training objective :: distillation objective.