add official Proteina-Complexa asset: README.md

README.md

@@ -0,0 +1,31 @@
+# Proteina-Complexa checkpoint mirror
+
+These are byte-for-byte copies of the six inference checkpoint files published
+by NVIDIA for Proteina-Complexa 160M v1. Atomic selects a matched flow-model and
+autoencoder pair for the `protein`, `ligand`, or `ame` backend variant.
+
+| Variant | Files | Official source | Upstream revision |
+| --- | --- | --- | --- |
+| Protein binder | `complexa.ckpt`, `complexa_ae.ckpt` | `nvidia/NV-Proteina-Complexa-Protein-Target-160M-v1` | `ffed199e32612b98ffa04f4640d34d37b137fca5` |
+| Ligand binder | `complexa_ligand.ckpt`, `complexa_ligand_ae.ckpt` | `nvidia/NV-Proteina-Complexa-Ligand-Target-160M-v1` | `bc90c8b2c701ceb52d5faef72600b6b5be880244` |
+| AME motif scaffolding | `complexa_ame.ckpt`, `complexa_ame_ae.ckpt` | `nvidia/NV-Proteina-Complexa-AME-160M-v1` | `9743d749a8754080a32fda857d95579dfa4dabae` |
+
+The files retain NVIDIA's names and are not converted, sharded, fine-tuned, or
+otherwise modified. `SHA256SUMS` is generated after the mirror download and
+must match the official Hub LFS digests recorded there.
+
+## License and notice
+
+The weights are governed by the NVIDIA Open Model License Agreement. Redistribu-
+tion requires both the complete agreement and the exact attribution notice;
+they are included here as `LICENSE-NVIDIA-OPEN-MODEL.txt` and `NOTICE`.
+Use must also comply with NVIDIA's Trustworthy AI terms and applicable law. The
+upstream model card is part of the source distribution:
+
+- https://huggingface.co/nvidia/NV-Proteina-Complexa-Protein-Target-160M-v1
+- https://huggingface.co/nvidia/NV-Proteina-Complexa-Ligand-Target-160M-v1
+- https://huggingface.co/nvidia/NV-Proteina-Complexa-AME-160M-v1
+
+Atomic supplies only a checkpoint mirror and a native-runtime bridge. Install
+and configure the upstream Proteina-Complexa/BioNeMo runtime separately before
+generating structures.