Representation shift, the L2 change in token embeddings through an MLP layer, is introduced as a training-free token importance metric compatible with FlashAttention.
The pagerank citation ranking: bringing order to the web
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Representation Shift: Unifying Token Compression with FlashAttention
Representation shift, the L2 change in token embeddings through an MLP layer, is introduced as a training-free token importance metric compatible with FlashAttention.