MoS: Unleashing parameter efficiency of low-rank adaptation with mixture of shards
A paper at ICLR 2025 presenting MoS, a method for more parameter-efficient LoRA through mixture of shards.
Forewarned is Forearmed: Harnessing LLMs for Data Synthesis via Failure-Induced Exploration
ICLR 2025 paper on data synthesis through failure-induced exploration.
qintong-li