1 paper · 1 filter
Mantas Lukauskas
Extractive prompt compression promises to cut LLM inference costs by removing low-information tokens, and learned compressors such as LLMLingua-2 report strong results on English b…