Hi, thanks for releasing the benchmark audit artifacts and configs.
I’m interested in reproducing the 0.09B DINO+MLP probe results on LIBERO. I noticed that the current shortcut_solvability release includes the configs and evaluation results, while the training/evaluation source code and checkpoints are intentionally excluded.
Do you have any plans to publicly release the training/evaluation code and, if possible, the corresponding checkpoints for the LIBERO DINO+MLP probe?
In particular, having the model implementation, data preprocessing/training pipeline, task-ID conditioning logic, and evaluation script would be very helpful for reproducing the reported results.
Thanks!
Hi, thanks for releasing the benchmark audit artifacts and configs.
I’m interested in reproducing the 0.09B DINO+MLP probe results on LIBERO. I noticed that the current shortcut_solvability release includes the configs and evaluation results, while the training/evaluation source code and checkpoints are intentionally excluded.
Do you have any plans to publicly release the training/evaluation code and, if possible, the corresponding checkpoints for the LIBERO DINO+MLP probe?
In particular, having the model implementation, data preprocessing/training pipeline, task-ID conditioning logic, and evaluation script would be very helpful for reproducing the reported results.
Thanks!