Transformer models encode concepts in quiet spectral regions, syntax in high-variance ones

Researchers have identified a dual geometry within transformer representations, where concept directions anti-concentrate in the spectral tail while static unembedding-row contrasts concentrate in high-variance directions. This phenomenon was observed across 17 models and 4 language pairs, with further evidence from SAE features and linear probes on Gemma and Llama. The findings suggest that transformers may move semantic content to spectrally quiet regions during processing, allowing concepts to be manipulated with less grammatical interference. AI

影响 Identifies a potential mechanism for how transformers process and store semantic information, which could inform future model architectures.

排序理由 This is a research paper published on arXiv detailing novel findings about transformer representations. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

• Gemma
• Llama
• Qwen 2.5

• 论文
• 其他

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →