scaling-dora-factored-norms

【效率与协作】Optimize adapter parameter efficiency at scale by decomposing row-wise norm computation into base/cross/BA components (15× memory reduction) and fusing kernel operations. Achieves 1.5–2.0× inference speedup with 77 GB peak VRAM reduction across 8–32B vision-language models; applies when training adapter-based models with strict memory budgets across …

独立 GitHub 快照 · 自动分类 · 以原作者文档为准