Tag
ProtStructQA introduces an executable benchmark for protein structural question answering that compiles natural language queries into a typed DSL program to evaluate LLMs on precise 3D measurements, revealing a capability-dependent denotation threshold where chain-of-thought becomes strongly beneficial above a certain model scale.
Introduces GHI, a Graphormer-over-conditioned-hypergraph-incidence framework for aspect-based sentiment analysis that represents linguistic evidence as token–hyperedge incidence relations, achieving state-of-the-art results on six benchmarks with only 247M parameters.
Atlarix is a desktop environment that pre-parses codebases into a node/edge graph, allowing coding agents to navigate architecture via queries instead of reading raw text, which improves performance of smaller local models.