单机多卡微调训练异常：deepspeed 模式 ，且最新tags 0.9.3版本：Unknown command: --local_rank=0/1

[angel1288]

Reminder

• I have read the above rules and searched the existing issues.

System Info

请问大家有没有遇到以下异常呢？
启动：deepspeed --num_gpus=2 llamafactory-cli train --deepspeed ds_z3_config.json .....
异常：无法正常识别自动产生的 --local-rank

Reproduction

deepspeed --num_gpus=2 llamafactory-cli train --deepspeed ds_z3_config.json .....

Others

No response

[hiyouga]

deepspeed --num_gpus=2 src/train.py

[angel1288]

deepspeed --num_gpus=2 src/train.py

@hiyouga 您好，又出现了新的问题呢：accelerate==1.5.0, AttributeError: module 'torch' has no attribute 'get_default_device'

[hiyouga]

Upgrade pytorch>=2.5
huggingface/transformers#38376

[angel1288]

deepspeed --num_gpus=2 src/train.py

@hiyouga 您好，又出现了新的问题呢：accelerate==1.5.0, AttributeError: module 'torch' has no attribute 'get_default_device'

已解决，降低版本：accelerate==1.4.0