1 dataset found

Groups: Large Language Models

Filter Results

  • CMR Scaling Law

    The dataset used in the paper is a mixture of general corpus and domain-specific corpus, with a power-law relationship between loss, mixture ratio, and training tokens scale.