【Savanna:为卷积多混合模型(StripedHyena 2)预训练提供强大基础设施。亮点:1. 支持大规模分布式训练,优化千卡集群性能;2. 提供多种优化技术,如a2a和p2p上下文并行化;3. 已成功训练多个模型,如StripedHyena 7B和Evo 2 40B,覆盖超9T tokens】
'Savanna: Pretraining infrastructure for research and application of convolutional multi-hybrid models (StripedHyena 2).'
GitHub: github.com/Zymrael/savanna
#深度学习# #预训练模型# #大规模训练# #AI创造营#
'Savanna: Pretraining infrastructure for research and application of convolutional multi-hybrid models (StripedHyena 2).'
GitHub: github.com/Zymrael/savanna
#深度学习# #预训练模型# #大规模训练# #AI创造营#