1
Continual LLM Upcycling: A Predictor-Gated Bank-Wise Sparsity Training Recipe for Dense-to-Sparse LLMs
一种从稠密到稀疏的持续训练新方法，利用预测器门控实现高效稀疏化，针对大语言模型场景。
arXiv:2606.10722v1 Announce Type: new Abstract: We study dense-to-sparse continual training as a way to construct channel-sparse large language models…