AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Shuffle the Context: RoPE-Perturbed Self-Distillation for Long-Context Adaptation

Zichong Li, Chen Liang, Liliang Ren, Tuo Zhao, yelong shen, Weizhu Chen

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 0 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
add RoPE-perturbed self-distillation (skip)Llama-3-8B-Instruct long-context SFT · RULER 64Kaccuracy (%): 57.25 -> 69.29Improved
add RoPE-perturbed self-distillation (skip)Qwen-3-4B · RULER 256Kaccuracy (%): 67.01 -> 68.10Improved
use reverse KL instead of forward KL (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): 57.8 -> 59.2Improved
use KL distillation instead of CLM on perturbed view (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): 56.5 -> 59.2Improved
sample random skip (s,y) instead of a fixed skip (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): 55.1 -> 59.2Improved
add dropout-based two-pass consistency instead of RoPE perturbation (ablation)generic regularizer · RULER 64K (Llama)accuracy (%): 47.9 -> 48.9No clear effect
add Gaussian attention noise consistency (ablation)generic regularizer · RULER 64K (Llama)accuracy (%): 47.9 -> 49.6No clear effect
remove positional encoding in perturbed view (NoPE) (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): 59.2 -> 29.5Hurt
use chunked permutation of index blocks (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): 47.9 -> 46.2Hurt
use random dilation of RoPE indices (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): 47.9 -> 53.4Improved
lower KL coefficient lambda to 0.2 (ablation)self-distillation objective · RULER 64K (Llama)accuracy (%): lambda 0.2: 54.0; 0.5: 60.0; 1: 59.2; 2: 58.9Mixed
combine skip self-distillation with LongCEQwen-3-4B after SFT · LongBench-v2accuracy (%): 29.40 -> 32.22Improved

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