intelligent-machine-learning/dlrover

Flash checkpoint does not support safetensors

Closed this issue · 2 comments

首先感谢dlrover团队的工作!

        if self.args.save_safetensors:
            logger.warn(
                "Flash checkpoint does not support safatensors "
                "and torch.save is used."
            )

请问flashckpt不支持保存safetensors格式的模型文件的原因是啥呢,以及后续会支持吗

There's no special reason for this. If there is a requirement, you're welcome to collaborate on building related capabilities together.