fsdp+sft error

[AriesJin]

Describe the bug
What the bug is, and how to reproduce, better with screenshots(描述bug以及复现过程，最好有截图)

# 80GiB * 2
nproc_per_node=8

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
accelerate launch --config_file "./examples/train/multi-gpu/fsdp_qlora/fsdp_offload.json" \
    swift/cli/sft.py \
    --model Qwen/Qwen2.5-VL-7B-Instruct \
    --train_type full \
    --dataset test/vlm.jsonl \
    --load_from_cache_file true \
    --split_dataset_ratio 0 \
    --torch_dtype bfloat16 \
    --attn_impl flash_attn \
    --freeze_vit true \
    --freeze_llm true \
    --freeze_aligner false \
    --num_train_epochs 3 \
    --per_device_train_batch_size 1 \
    --learning_rate 5e-6 \
    --gradient_accumulation_steps 2 \
    --eval_steps -1 \
    --save_steps 1000 \
    --save_total_limit 10 \
    --logging_steps 5 \
    --max_length 3072 \
    --output_dir output \
    --warmup_ratio 0.05 \
    --dataloader_num_workers 4 \
    --dataset_num_proc 8 \
    --padding_free true \
    --gradient_checkpointing true 

Your hardware and system info
Write your system info like CUDA version/system/GPU/torch version here(在这里给出硬件信息和系统信息，如CUDA版本，系统，GPU型号和torch版本等)

torch                         2.6.0+cu124
torchao                       0.13.0
torchaudio                    2.6.0+cu124
torchelastic                  0.2.2
torchvision                   0.21.0+cu124
tqdm                          4.67.1
traitlets                     5.14.3
transformers                  4.57.0
transformers-stream-generator 0.0.5

Additional context
Add any other context about the problem here(在这里补充其他信息)

[slin000111]

Describe the bug What the bug is, and how to reproduce, better with screenshots(描述bug以及复现过程，最好有截图)

# 80GiB * 2
nproc_per_node=8

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
accelerate launch --config_file "./examples/train/multi-gpu/fsdp_qlora/fsdp_offload.json" \
    swift/cli/sft.py \
    --model Qwen/Qwen2.5-VL-7B-Instruct \
    --train_type full \
    --dataset test/vlm.jsonl \
    --load_from_cache_file true \
    --split_dataset_ratio 0 \
    --torch_dtype bfloat16 \
    --attn_impl flash_attn \
    --freeze_vit true \
    --freeze_llm true \
    --freeze_aligner false \
    --num_train_epochs 3 \
    --per_device_train_batch_size 1 \
    --learning_rate 5e-6 \
    --gradient_accumulation_steps 2 \
    --eval_steps -1 \
    --save_steps 1000 \
    --save_total_limit 10 \
    --logging_steps 5 \
    --max_length 3072 \
    --output_dir output \
    --warmup_ratio 0.05 \
    --dataloader_num_workers 4 \
    --dataset_num_proc 8 \
    --padding_free true \
    --gradient_checkpointing true 

**Your hardware and system info** Write your system info like CUDA version/system/GPU/torch version here(在这里给出硬件信息和系统信息，如CUDA版本，系统，GPU型号和torch版本等)

torch                         2.6.0+cu124
torchao                       0.13.0
torchaudio                    2.6.0+cu124
torchelastic                  0.2.2
torchvision                   0.21.0+cu124
tqdm                          4.67.1
traitlets                     5.14.3
transformers                  4.57.0
transformers-stream-generator 0.0.5

Additional context Add any other context about the problem here(在这里补充其他信息)

accelerate版本看下是哪个