AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

SkillFactory: Self-Distillation for Learning Cognitive Behaviors

Zayne Rea Sprague, Jack Lu, Manya Wadhwa, Sedrick Keh, Mengye Ren, Greg Durrett

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use SkillFactory SFT init before GRPO instead of RL onlyQwen2.5-1.5B-Instruct · Countdown 4-6argaccuracy: 15.8 -> 25.1Improved
SkillFactory SFT init vs R1 distillation init before GRPOQwen2.5-1.5B-Instruct · Countdown 4-6argaccuracy: 21.2 -> 25.1Improved
SkillFactory SFT alone has lower accuracy than R1 distillation SFT but better post-RLQwen2.5-1.5B-Instruct · Countdown 4-6argaccuracy: SFT 2.8 vs 11.7; post-RL 25.1 vs 21.2Mixed
STaR self-distillation init before GRPOQwen2.5-1.5B-Instruct · Countdown 4-6argaccuracy: 15.8 -> 9.7Hurt
SkillFactory init vs RL only on out-of-domain averageQwen2.5-1.5B-Instruct · OOD tasks overalloverall accuracy: 31.9 -> 35.7Improved
SkillFactory init vs RL only on OOD average with larger modelsQwen2.5-7B-Instruct · OOD tasks overalloverall accuracy: 32.9 -> 54.1 (R1 Distill -> GRPO: 70.1)Improved
add reflections to silver traces (ablation)SkillFactory · OOD tasks overall (ablation eval)overall accuracy: 23.3 -> 32.0Improved
order samples so traces end on a correct sample (ablation)SkillFactory · OOD tasks overall (ablation eval)overall accuracy: 24.1 -> 32.0Improved
use diverse solve prompts rather than a single prompt (ablation)SkillFactory · OOD tasks overall (ablation eval)overall accuracy: 29.0 -> 32.0Improved
add SFT on structured traces instead of only an instruction prompt (ablation)RL on Qwen2.5-1.5B-Instruct · OOD tasks overall (ablation eval)overall accuracy: 25.8 -> 32.0Improved
scale SkillFactory SFT data from 1k to 10k rows (ablation)Qwen2.5-7B-Instruct · OpenThoughts math/science overalloverall accuracy: 42.1 -> 40.6No clear effect
SkillFactory vs QwQ distillation at 1k rowsQwen2.5-7B-Instruct · OpenThoughts math/science overalloverall accuracy: 33.5 -> 42.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy