AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

One LR Doesn’t Fit All: Heavy-Tail Guided Layerwise Learning Rates for LLMs

Di He, Songjun Tu, Keyu Wang, Lu Yin, Shiwei Liu

ICML 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 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 Hill-alpha guided layerwise LR (LLR)AdamW LLaMa pretraining · FineWeb 60Mvalidation perplexity: 21.94 -> 20.30Improved
add LLRAdamW LLaMa-1B · FineWeb 1Bvalidation perplexity: 9.77 -> 9.59Improved
add LLRLLaMa-1B · 7 zero-shot commonsense tasksaverage accuracy (%): 47.09 -> 49.02Improved
add LLR at larger token budget (3B model, 30B tokens)LLaMa-3B · zero-shot commonsenseaverage accuracy (%): 48.58 -> 50.61Improved
add LLR at 5x chinchilla tokens (100B)LLaMa-1B · zero-shot commonsenseaverage accuracy (%): 51.16 -> 51.86Improved
add LLRLlama-3.2-1B finetuning · Commonsense reasoning finetuningaverage accuracy (%): 52.94 -> 53.58Improved
add LLR to ViT-TinyViT-Tiny · ImageNet-1Ktop-1 accuracy (%): 67.42 -> 68.51Improved
add LLR with Muon optimizerLLaMa-350M 20B tokens · zero-shot averageaverage accuracy (%): 45.53 -> 47.08Improved
add LLR with AdamW at 3x chinchillaLLaMa-350M 20B tokens · zero-shot averageaverage accuracy (%): 44.50 -> 46.03Improved
invert the Hill-alpha to LR mapping (LLR-inv) instead of the proposed one (ablation)LLR · FineWeb 135Mvalidation perplexity: 17.03 -> 27.18Hurt
use linear instead of Log2 or Sqrt LR assignment (ablation)LLR · LLaMa-60M, LR 0.0005validation perplexity: 17.86 (Log2) -> 17.55 (Linear)Improved
fix embedding LR at upper bound instead of LLR-unembed (ablation)LLR · LLaMa pretrainingvalidation perplexityImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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