Reproduce results pi0-primitive-10task

[zehao-wang]

Hi, thanks for providing the benchmark, but I met issues of reproducing the results in pi0-primitive-10task. The README file seems not correct, I tried to implement the lora version and finetuning the model pi0-base from Physical-Intelligence on vlabench lerobot. The evaluation results is close to 0 for track1.

Could you please provide:

1. The config of the finetuning model, is it a lora version or full finetuning version?
2. The finetuning is happening on which pi0-base model, the one provided in openpi official or you have another pretrained version

Thanks.