GRKD replaces absolute probability matching with a pairwise ranking loss on teacher-ordered pairs and reports improved LLM distillation results.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Group Relative Knowledge Distillation: Learning from Teacher's Relational Inductive Bias
GRKD replaces absolute probability matching with a pairwise ranking loss on teacher-ordered pairs and reports improved LLM distillation results.