ms-swfit 微调qewn2.5-vl 报错 IndexError: index 4247 is out of bounds for dimension 0 with size 4247

[Oukaishen]

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)

SFT qwen2.5-vl在某些corner-case情况下报错

swift sft
--model XXXX/qwen/Qwen2.5-VL-3B-Instruct
--dataset XXX/train.json
--train_type full
--torch_dtype bfloat16
--truncation_strategy right
--deepspeed zero3
--freeze_vit true
--num_train_epochs 1
--per_device_train_batch_size 1
--per_device_eval_batch_size 1
--learning_rate 1e-5
--gradient_accumulation_steps 4
--max_new_tokens 32
--eval_steps 2000
--save_steps 2000
--save_total_limit 10
--logging_steps 10
--max_length 4096
--output_dir output
--warmup_ratio 0.05
--dataloader_num_workers $GPUS_PER_NODE

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)

GPU - H20 X 8

ms_swift 3.3.0.post1
transformers 4.49.0
transformers-stream-generator 0.0.5
nvidia-cuda-cupti-cu12 12.4.127
nvidia-cuda-nvrtc-cu12 12.4.127
nvidia-cuda-runtime-cu12 12.4.127
torch 2.5.1
torchvision 0.20.1
qwen-vl-utils 0.0.10

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

IndexError: index 4247 is out of bounds for dimension 0 with size 4247

[Jintao-Huang]

please remove --truncation_strategy right