SinkTrack anchors LLMs to initial context by modifying the attention sink token with injected features, yielding gains on textual and multimodal tasks.
"" h_ori: hidden states of the original sequence (Lori ×D h)) h_info: hidden states of the external information (L inf o ×D h)) cfg: configuration for injection rules
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SinkTrack: Attention Sink based Context Anchoring for Large Language Models
SinkTrack anchors LLMs to initial context by modifying the attention sink token with injected features, yielding gains on textual and multimodal tasks.