A September 2, 2026 post describes NoRA, which normalizes the columns of LoRA matrix A at initialization, and reports higher average results than several adaptation methods in SFT and RLVR.
Published on September 2, 2026, the post introduces Normalized Low-Rank Adaptation (NoRA), an approach that normalizes the columns of LoRA matrix A during initialization. According to the post, NoRA-init achieves most of the gains without keeping normalization active throughout training.
The post reports higher averages than several adaptation methods in SFT and RLVR, without providing values or experimental conditions here. The proposal may interest readers studying LoRA training; consult the original post to check its methods, metrics, and limits. If you use AI to analyze the material, avoid submitting your organization's internal or personal data.