code

Visual perspective-taking benchmark + Isaac Lab data pipeline. Synthetic VPTnav dataset for testing line-of-sight reasoning in vision models.

Python

SAE feature steering on Gemma 3 — bend a language model's emotional weather via Gemma Scope 2 sparse autoencoders.

Python

Evaluation metrics for physion-style physical reasoning benchmarks.

Python

Location-Specific RNA-Protein Interaction Transformer — predicts residue-level RPIs from primary sequence alone.

Python

LLM-driven molecule and materials generation with logical feedback loops. Drug candidates + low-k dielectrics.

Python

Open-source library for deep learning in drug discovery, materials science, and computational biology. 40k+ users. Contributing models, infrastructure, tutorials.

Python