AI RESEARCH

Forecasting Downstream Performance of LLMs With Proxy Metrics

arXiv CS.LG

ArXi:2605.18607v1 Announce Type: cross Progress in language model development is often driven by comparative decisions: which architecture to adopt, which pre