GateKD is a confidence-gated closed-loop distillation framework that improves multi-step reasoning transfer from LLMs to smaller models by dynamically filtering supervision based on teacher reliability.
arXiv preprint, arXiv:2405.19842
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
GateKD: Confidence-Gated Closed-Loop Distillation for Robust Reasoning
GateKD is a confidence-gated closed-loop distillation framework that improves multi-step reasoning transfer from LLMs to smaller models by dynamically filtering supervision based on teacher reliability.