Accurate perception of both dynamic road participants and static road infrastructure is critical for the reliable operation of autonomous vehicles (AV). Traditionally, models in this field have relied on supervised learning, but our project aims to leverage recent advancements in semi- and self-supervised learning to enhance AV perception capabilities. To achieve this, we require not only labeled datasets but also a substantial amount of unlabeled data, which is essential for training and improving our mapping algorithms. This necessitates additional storage resources to accommodate the large-scale datasets required for effective learning. Furthermore, our approach involves distilling knowledge from large foundation models in the vision/camera domain to the lidar and radar sensor domains. To facilitate this transfer, we need to cache features extracted from these models. Performing inference on such large models during training significantly slows down the process, making feature caching essential for efficiency. Given these requirements, we request additional storage capacity to support our research efforts effectively. This will enable faster training iterations, better utilization of semi/self-supervised learning techniques, and improved performance in perception for autonomous driving. We also need to store checkpoints and debug information to gain knowledge how the learning within the models occur.