ThinkTuning uses teacher feedback grafted onto GRPO rollouts to teach small language models reflective reasoning, improving MATH-500, AIME, and GPQA-Diamond accuracy over vanilla GRPO.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ThinkTuning: Instilling Cognitive Reflections without Distillation
ThinkTuning uses teacher feedback grafted onto GRPO rollouts to teach small language models reflective reasoning, improving MATH-500, AIME, and GPQA-Diamond accuracy over vanilla GRPO.