AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

Boomerang Distillation Enables Zero-Shot Model Size Interpolation

Sara Kangaslahti, Nihal V. Nayak, Jonathan Geuter, Marco Fumero, Francesco Locatello, David Alvarez-Melis

ICLR 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Few-shot and metric distillation · 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
initialize student from teacher weights instead of randomly (ablation)Boomerang distillation, Qwen3-4B-Base · 10 classification and 3 generation tasksaccuracy vs parameter countImproved
add knowledge distillation training to the layer-pruned student (versus naive layer pruning) (ablation)Boomerang distillation, Qwen3-4B-Base · classification and generation tasksaccuracyImproved
add per-layer cosine alignment loss to CE+KL (ablation)Boomerang distillation, Qwen3-4B-Base · WikiTextperplexity of interpolated modelsImproved
add per-layer alignment loss (ablation)Boomerang distillation, Qwen3-4B-Base · downstream classification/generation, majority of sizesaccuracyNo clear effect
add per-layer alignment loss (ablation)Boomerang distillation, Qwen3-4B-Base · extreme intermediate sizes (first/last layers patched)stability of accuracyImproved
patch teacher layers into the student (boomerang) versus layer pruning LaCo and ShortGPTQwen3-4B-Base interpolated models · classification and generation tasksaccuracyImproved
compare interpolated models to standard distilled models of same sizeBoomerang distillation, Qwen3-4B-Base · classification and generation tasksaccuracyMixed
compare interpolated models to standard distilled models on Pythia-2.8BBoomerang distillation, Pythia-2.8B · classification tasksaccuracyHurt
use student with every 3rd layer kept instead of every 2nd (ablation)Boomerang distillation, Qwen3-4B-Base · student size ablationinterpolation qualityNo clear effect
keep every 4th or 5th layer for student init instead of every 2nd (ablation)Boomerang distillation, Qwen3-4B-Base · student size ablationinterpolation smoothnessHurt
choose student layers to preserve first/last cosine similarity in each teacher block (ablation)Boomerang distillation, Qwen3-4B-Base · small student modelsinterpolation smoothness: 505M parameters, 8.7x compressionImproved
increase student training tokens from 0.5B up to 3.1B (ablation)Boomerang distillation, Qwen3-4B-Base · token budget ablationinterpolated model performance: 0.5B, 1B, 1.5B, 2B, 2.5B, 3.1B tokensImproved

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