datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
rollout_stackcups_bcq_q2_r2rollout_VarunGiridhar3_abcdit_stackcups_right_filteredbc_r1_005000_r0_20260809_000332rollout_VarunGiridhar3_abcdit_stackcups_right_005000_bcq_negft010k_r0_20260808_194142rollout_stackcups_bcq_q1_r1rollout_abcdit_stackcups_right_005000_bc_r0_20260808_165837rollout_stackcups_filteredbc_r2_r0stack-cups-rollouts-decoded-20260923
Decoded rollouts, 22-23 September 2026 night session
One rollout_NNNN_decoded.npz per rollout, 0276 through 0323 (48 runs, 23:00:23
to 00:58:39), plus MANIFEST.json with per-array quality statistics.
Produced by ~/yiming/tools/decode_rollouts/decode_rollout.py. Source rollouts
are untouched in ~/yiming/workstation_client/rollouts/rollout_NNNN/.
What was actually decoded
The vae_* arrays in a source rollout are latents, shape (1, 48, T, 10, 20),
not images. The VAE… See the full description on the dataset page: https://huggingface.co/datasets/chillinfree333/stack-cups-rollouts-decoded-20260923.