AI Paper Atlas

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

Pre-training LLM without Learning Rate Decay Enhances Supervised Fine-Tuning

Kazuki Yano, Shun Kiyono, Sosuke Kobayashi, Sho Takase, Jun Suzuki

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 4 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
remove LR decay in pre-training (WSO instead of best decay scheduler)Llama-3 1B · after SFTSFT Task Avg: +0.3 vs best decayImproved
remove LR decay in pre-training (WSO)Llama-3 8B · after SFTSFT Task Avg: +1.1 vs best decayImproved
remove LR decay in pre-training (WSO)Llama-3 1B and 8B · after pre-trainingPT valid loss and PT Task Avg: loss +0.071 (1B), +0.127 (8B); task avg -1.7 (1B), -0.8 (8B)Hurt
decay LR to zero instead of 10% (alpha 0.1 to 0.0) in WSD (ablation)Llama-3 1B · after SFTSFT Task Avg: +0.0 -> -1.0Hurt
use constant LR in mid-training instead of linear decay (alpha_mid 1.0 vs 0.0) (ablation)WSD pre 0.1, 1B · after SFTSFT Task Avg: -1.3 -> -0.5Improved
use constant LR through pre- and mid-training (WSO) in three-stage pipelineLlama-3 8B · after SFTSFT Task Avg: +1.1 vs best decayImproved
use WSO in over-training with 2T tokensLlama-3 1B · after SFTSFT Task Avg: +0.7Improved
use WSO in over-training with mid-trainingLlama-3 1B · after SFTSFT Task Avg: +1.4Improved
remove LR decay (WSO) to reduce sharpnessLlama-3 1B · sharpness of pre-trained modelHessian trace: decay models 2-3x higher sharpnessImproved
higher sharpness of pre-trained model (ablation)Llama-3 1B · SFT performanceSFT score: Pearson r = -0.709Hurt

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