each episodes faild when pi05_ft_vlabench_primitive

[JuilieZ]

Hi,thx for your such wonderful work and provide the checkpoints!
but I met some issues when I run the eval script.
I use the models 'VLABench/pi05-primitive-10task' you released at huggingface ,and I follow your instructions ,bash eval.sh but got poor performance
I guess I made some mistakes but I cannot figure it out.Could you help me plz?

0_success_False_progress_0.00.mp4