Masked Curiosity Checkpoints\n\nThis repository contains the model checkpoints for the Masked Curiosity project, specifically the selected 20 datasets training run.\n\n## Checkpoints\n\n- consolidated_epoch_1.pth: Consolidated checkpoint from epoch 1\n- zero_to_fp32.py: Utility script to convert DeepSpeed ZeRO checkpoints to FP32 format\n\n## Usage\n\nThese checkpoints can be loaded using the model architecture defined in the main repository. The consolidated checkpoint is in FP32 format and can be loaded directly. For the individual epoch checkpoints, use the zero_to_fp32.py script to convert them from DeepSpeed ZeRO format to FP32 format.\n\n## Model Architecture\n\nThe model architecture is defined in model_large.py in the main repository. It is a co-training model that combines visual and proprioceptive inputs using a masked autoencoder approach.
