# MolmoER / Molmo2-ER training corpus

From the catalogue. Listed by one researcher from one source page.

- id: `molmoer-molmo2-er-training-corpus`
- kind: language and reasoning annotation
- robot or device: mixed: RoboVQA, RoboPoint, RefSpatial, VSI-590K, SenseNova-SI, VST-P, SAT, SIMS-VSI, CLEVR (class: cross embodiment)
- how collected: re-packaged existing spatial and embodied QA sets
- size as stated: 3.3M-sample corpus
- hours: not stated | hours only claimed: none | episodes: not stated
- year: 2026
- organisation: Allen Institute for AI
- licence: per component (apache-2.0, cc-by-4.0, mit, cc-by-nc-4.0) (class: non commercial)
- access: open (class: open)
- link (dataset): https://huggingface.co/datasets/allenai/Molmo2-ER-VST-P
- Hugging Face repository: `allenai/Molmo2-ER-VST-P`, 4,604 downloads in 30 days, https://huggingface.co/datasets/allenai/Molmo2-ER-VST-P
- note: Nine allenai/Molmo2-ER-* repos, one per source dataset.

Confirm the licence at the link before relying on it.