Co-Training Visual Checkpoints\n\nThis repository contains various model checkpoints from the Masked Curiosity project, organized by experiment type.\n\n## Structure\n\n- checkpoints/: Base model checkpoints\n- checkpoints-no-tube-proprio/: Checkpoints without tube proprioception\n- checkpoints_sim_good_data/: Checkpoints trained on good simulation data\n- checkpoints_sim_trash_data/: Checkpoints trained on trash simulation data\n- checkpoints_sim_trash_data_proprio_augment/: Checkpoints trained on trash simulation data with proprioception augmentation\n- checkpoints-with-all-feature-small/: Checkpoints with all features (small model)\n- checkpoints-without-temporal_stride/: Checkpoints without temporal stride\n\nEach directory contains the model checkpoints for that specific experiment configuration. The checkpoints are stored in DeepSpeed ZeRO format and can be converted to FP32 using the zero_to_fp32.py script.
