昇腾卡训练不支持offload

hiyouga / LLaMA-Factory

Efficiently Fine-Tune 100+ LLMs in WebUI (ACL 2024)

https://arxiv.org/abs/2403.13372

Apache License 2.0

31.82k stars 3.91k forks source link

Closed wangbing35 closed 2 months ago

wangbing35 commented 3 months ago

使用昇腾卡训练大参数量模型，deepspeed stage 3 + offload模式下，提示RuntimeError: inplace tensor self must be NPU-Tensor

ds_z3_offload_config.json

No response

No response

huyz-git commented 3 months ago

https://github.com/microsoft/DeepSpeed/issues/5585 等 deepspeed 下一个 release，或者自己用 master 代码编译 deepspeed 。

wangbing35 commented 2 months ago

deepspeed升级到0.14.3，支持offload了