AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Stabilizing Native Low-Rank LLM Pretraining

Paul Janson, Edouard Oyallon, Eugene Belilovsky

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws

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 spectral renormalization on top of orthogonalization (ablation)Spectron · FineWeb val (Factorized Llama-94M)perplexity: 20.95 -> 20.26Improved
add orthogonalization on top of spectral renormalization (ablation)Spectron · FineWeb val (Factorized Llama-94M)perplexity: 24.01 -> 20.26Improved
add either spectral control component to naive factorized training (ablation)naive low-rank training · FineWeb val (Factorized Llama-94M)perplexity: 1042.02 -> 24.01Improved
use rank ratio 0.4 instead of 0.25 (ablation)Spectron · FineWeb val (Factorized Llama-94M)perplexity: 20.26 -> 20.00No clear effect
reduce rank ratio to 0.125 (ablation)Spectron · FineWeb val (Factorized Llama-94M)perplexity: 20.26 -> 42.65Hurt
use Spectron instead of self-guided training with dense auxiliary weightsFactorized Llama · FineWeb val (Factorized Llama-454M)perplexity: 13.70 -> 12.11Improved
use Spectron instead of naive AdamWFactorized Llama · FineWeb val (Factorized Llama-94M)perplexity: 26.43 -> 21.86Improved
train factorized model with Spectron at higher rank ratio (ablation)Factorized Llama-94M · FineWeb val loss (rank ratio 0.4)validation loss: 3.070 (dense) -> 3.028Improved
train Spectron factorized model at rank ratio 0.25 versus dense (ablation)Factorized Llama-94M · FineWeb val loss (rank ratio 0.25)validation loss: 3.070 (dense) -> 3.084No clear effect
combine Spectron with CoLA architectureCoLA · FineWeb val (Factorized Llama-94M)perplexity: 24.21 -> 19.95Improved
use Spectron instead of LoROlow-rank training · FineWeb val (Factorized Llama-94M)perplexity: 24.56 -> 21.86Improved
use Muon with self-guided training instead of AdamWSelf-guided training · FineWeb val (Factorized Llama-94M)validation lossHurt

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