EndPrompt induces long-context generalization in LLaMA models via a two-segment short-sequence construction with terminal positional anchoring, outperforming full fine-tuning and prior methods on RULER and LongBench while using less compute.
Smith, and Hajishirzi Hannaneh
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
No citing papers match the current filters.