借助 DeepSpeed 和 FairScale,通过 ZeRO 容纳更大模型并加速训练
Fit More and Train Faster With ZeRO via DeepSpeed and FairScale
精选理由
介绍如何借助ZeRO与DeepSpeed、FairScale降低显存占用,加速大模型训练。
Fit More and Train Faster With ZeRO via DeepSpeed and FairScale
介绍如何借助ZeRO与DeepSpeed、FairScale降低显存占用,加速大模型训练。