New method predicts LLM performance without expensive testing

Original: Forecasting Downstream Performance of LLMs With Proxy Metrics

Source: arxiv.org ↗