Retry the destination list_repo_files with backoff (and fall back to an
empty resume set) so a single transient huggingface.co read timeout no
longer fails an entire SLURM rank and skips its dataset slice.
Add migrate_molmoact.py to migrate the SO-100/101 datasets listed by
allenai/MolmoAct2-SO100_101-Dataset (repo ids derived from language_annotations
folder names), skipping any already in lerobot/community_dataset_v3. Generalize
run_migration.migrate_one/_write_dataset_card with a backward-compatible
standalone mode (download a whole standalone repo, provenance -> source repo)
and extend slurm_migrate.py with --source molmoact / --reference-repo.
- download_subfolder: fetch only the target sub-dataset subtree instead of
enumerating the whole community_dataset_v3 monorepo tree (fixes apparent hang)
- normalized SO gripper (RANGE_0_100) left in native 0..100 frame, matching
degrees_new datasets, instead of remapping to +/-45deg
- uncalibrated normalized datasets: skip identity value rewrite, keep normalized
units and flag them APPROXIMATE on the dataset card
- remove --allow-uncalibrated flag and its CANON_IS_CALIBRATED side effect
CarolinePascal