internlm-xcomposer2_5-7b fine-tuning with multiple gpus reports an error "all tensors to be on the same device"

[thonglv21]

CUDA_VISIBLE_DEVICES=0,1 swift sft --model_type internlm-xcomposer2_5-7b-chat --dataset myData.jsonl

Expected all tensors to be on the same device, but found at least two devices

Looking forward to your answer!

[Jintao-Huang]

NPROC_PER_NODE=2 CUDA_VISIBLE_DEVICES=0,1 swift sft --model_type internlm-xcomposer2_5-7b-chat --dataset myData.jsonl --deepspeed default-zero2