Add data/eval - 5,850 policy-evaluation episodes with outcomes (510 GB)
#33
by yi-xdof - opened
Automated S3->HF mirror chunk.
Folders:
- load_the_plates_into_the_dish_rack
- throw_the_plastic_bottles_in_the_bin
- turn_the_mug_right_side_up
yi-xdof changed pull request status to open
yi-xdof changed pull request title from chunk 00: 3 folders, 0.51 TB to Add data/eval - 5,850 policy-evaluation episodes with outcomes (510 GB)
Consolidated onto main on 2026-09-18: all 11,700 files (5,850 episode.mcap + 5,850 eval_result.json, 510,057,811,121 bytes) copied from refs/pr/33 to data/eval/ in 4 linear commits.
Verification: 11,700/11,700 present on main, size-exact vs the S3 source; sidecar byte-identical; mcap head/tail byte-check passed; other data/ splits untouched.
Contents: real-robot policy-evaluation rollouts for 35 checkpoints across 3 tasks (load the plates into the dish rack 1,992; throw the plastic bottles in the bin 2,008; turn the mug right side up 1,850), each episode with an eval_result.json outcome sidecar (strict_success, task_progress, rubric scores, policy).
yi-xdof changed pull request status to closed