A distillation scaling law predicts student cross-entropy from teacher loss, student size, and data, and gives compute-optimal teacher-student allocations.
Our setting automatically satisfies consistency as there is no augmentation policy
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Distillation Scaling Laws
A distillation scaling law predicts student cross-entropy from teacher loss, student size, and data, and gives compute-optimal teacher-student allocations.