architecture · first seen 16 Oct 2026
unembedding matrix
Output projection mapping final hidden states to vocabulary logits in language models.
Mentions per week
PapersCodeDiscussion
Earliest sources
- What LLMs Do in the Shadows: Reading Context Reliance from the Unembedding Dark SpacePapers · 16 Oct 2026 · Alexandra Kuleshova, Andrey Volodichev, Daria Kotova, Alexey Zaytsev
Latest sources
- What LLMs Do in the Shadows: Reading Context Reliance from the Unembedding Dark SpacePapers · 16 Oct 2026 · Alexandra Kuleshova, Andrey Volodichev, Daria Kotova, Alexey Zaytsev