Datasetpublic

ocean-world/e04-aquatic-pusht

e04-results — LeWM toy-world-model results (E0.4) + action-separability probes Archived outputs of the E0.4 program (from-scratch ViT-tiny pixel world models on PushT-style toys ± aquatic disturbances — wet-PushT, current-PushT) plus the later action-separability / LDAD-ingredient probe inputs. ~320 files, mostly result JSONs, configs, and pod snapshots. Narrative + verdicts: ../E0.4-RESULTS.md → superseded by ../E0.4-FINAL-SUMMARY.md; distilled in… See the full description on the dataset page: https://huggingface.co/datasets/ocean-world/e04-aquatic-pusht.

sourceHugging Faceupdated 22d agoView on Hugging Face

0likes34downloads

Dataset Card

e04-results — LeWM toy-world-model results (E0.4) + action-separability probes

Archived outputs of the E0.4 program (from-scratch ViT-tiny pixel world models on PushT-style toys ± aquatic disturbances — wet-PushT, current-PushT) plus the later action-separability / LDAD-ingredient probe inputs. ~320 files, mostly result JSONs, configs, and pod snapshots.

—Narrative + verdicts: ../E0.4-RESULTS.md → superseded by ../E0.4-FINAL-SUMMARY.md; distilled in ../../../handoff/memory/ocean_world_e04_lewm_toys.md.
—e04-pod-authoritative/ — the copy of the run scripts marked authoritative at the time (the naming is the point: other copies are stale).
—action_separability/ — inputs/outputs of the LDAD ingredient probe (../FINDINGS_ldad_ingredient_probe_2026-07-19.md); the two big npz inputs are gitignored and staged in the HF media repo under analysis-artifacts/e04-action-separability/.
—Training/eval code that produced these lives in the le-wm clone — reproduce via ../../../vendor-patches/le-wm/.

Archive directory: don't edit in place; these back published numbers.