Jev-Spatial — Apache License, Version 2.0

Based on Molmo2-ER and Molmo2 by the Allen Institute for AI and contributors.
Model: https://huggingface.co/allenai/Molmo2-ER
Source: https://github.com/allenai/molmo2
Pinned revision: dab22564403d2607855bb1fffb0721285b445081
Upstream license: Apache License, Version 2.0 (retained in LICENSE).

The five native Python modules under spatial_jev/molmo2 are unmodified copies
of the model/processor implementation at the pinned revision. Jev-Spatial adds
a shared classification head, hierarchical inference, training and evaluation
code. The released weights merge the trained LoRA into Molmo2-ER and omit the
unused autoregressive vocabulary head. Configuration metadata is adapted to
load the implementation from the installed source package.

Jev / System One by TypeSafe is the project's conceptual inspiration.
This is an independent project, not an official TypeSafe implementation or
a reproduction of proprietary RLCD. Full acknowledgements are in README.md.

Training and benchmark media are not included. Their original attribution
and terms remain with their respective authors and distributors.