However, existing meth- ods typically rely on gradient computations, making them computationally prohibitive for billion-parameter models and precluding batch parallelization

For-Value: Efficient Forward-Only Data Valuation for finetuning LLMs, VLMs Wenlong Deng1⋆, Qi Zeng3, Jiaming Zhang1, Minghui Chen1, Zixin Ding3, Christos Thrampoulidis1,Boying Gong 3†,Xiaoxiao Li 1,2† 1University of British Columbia, 2Ve · 2023