问题描述:
调用代码
deepspeed_plugin = DeepSpeedPlugin(zero_stage=2, gradient_accumulation_steps=args.gradient_accumulation_steps)
报错如图所示:
根因定位:
链接:https://github.com/microsoft/DeepSpeed/issues/3228
This error only occurs when using deepspeed v0.9.0 and zero stage 2.
我升级了deepspeed v0.9.0到0.9.3
就没有报错了