LATO.2

From the project README

A Python process extension that exposes the complete pinned LATO.2 inference pipeline as four mesh workflow nodes in upstream Modly. License notice: the exact native profile installs NVIDIA nvdiffrast, whose NVIDIA Source Code License (1-Way Commercial), Section 3.3, limits the Work and derivatives to non-commercial research or evaluation. Read THIRDPARTYNOTICES.md before installation or use. DINOv2 archive notice: the selected DINOv2 register-token path is Apache-2.0, but the complete pinned source archive also contains imported Cell-DINO and X-Ray-DINO modules with separate terms, including noncommercial-research material. This extension does not download or select their checkpoints. See the DINOv2 section of THIRDPARTYNOTICES.md for the exact scope, license copies, and required X-Ray DINO notice.

This is a pre-hardware release candidate. The implementation targets upstream Modly 0.4.2 audited at commit 8d08249fbd4f2678d923c213a8cb2902d57611c9, but a full Install from GitHub → setup/Repair → UI workflow → real checkpoint generation has not yet been completed on the target GPUs. In particular, Windows/Turing and Linux ARM64 full-model parity remain unverified. The compatibility rows below describe implemented setup routes, not completed hardware certification.

The highest completed validation tier is the repository's automated contract/unit suite and strict extension validation on Linux ARM64 with CPython 3.12. This proves setup-contract selection and lock separation, not a real installation. Native compilation, real-GPU dependency setup, and checkpoint generation are the next tier and remain pending on every platform row below.