model card initial commit

README.md

@@ -1,3 +1,30 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+---
+## Dyno Psi Overview
+
+Dyno Psi-1 is a generative protein design model for *de novo* binder design.  
+
+[![GitHub](https://img.shields.io/badge/GitHub-dynopsi-F1E0CC?logo=github&logoColor=black)](https://github.com/dynotx/dynopsi) 
+
+
+
+### Description 
+
+The Dyno Psi approach to binder design consists of a backbone generation model (Dyno Psi-1), a sequence design component, and *in silico* filters. Dyno Psi-1 generates backbones via a flow-matching-based denoising model. Sequences are designed post hoc using an inverse folding model, and candidates are evaluated with structure-based filters. 
+
+This Hugging Face repository and corresponding [GitHub repository](https://github.com/dynotx/dynopsi) enable the use of the Dyno Psi-1 backbone generation model.
+
+Dyno Psi-1 was trained in multiple stages to improve its ability to generalize. Training began with large-scale monomer generation on structures from the PDB and AlphaFold Database (AFDB), followed by joint training on monomers and protein–protein interactions. To expand structural diversity beyond experimentally solved complexes, millions of synthetic intra-domain interaction pairs derived from AFDB monomers were generated and clustered to reduce redundancy. This mixture of experimental and synthetic interaction data enables the model to learn diverse geometric and interface features relevant to *de novo* binder generation.   
+
+
+### Architecture & Training
+ Dyno Psi-1 is a ~200M-parameter coordinate denoising model built on a scalable protein transformer with triangle updates, and has a maximum residue context window of 512.
+
+ #### Training Phases
+| Phase | Steps | GPUs | Max Tokens | Data |
+|-------|-------|------|------------|------|
+| Monomer pretraining | 232k | 16× H200 | 256 | PDB monomers, AFDB monomers |
+| Interface-aware pretraining | 143k | 96× H200 | 512 | PDB monomers, AFDB monomers, PDB multimers, ~ 4M synthetic intra-domain interaction pairs (~1M clusters) |
+| Binder design fine-tuning | 154k | 96× H200 | 512 | PDB complexes, AF2-filtered synthetic interaction pairs |
+