Tag
The paper introduces Braco, a lightweight token parameterization method for visual token compression in vision-language models, achieving significant efficiency gains and accuracy under extreme compression budgets.
A blog post shows that a formally verified Lean implementation of DEFLATE compression outperforms a pure-Rust implementation in both speed and compression ratio at typical levels. The author attributes this to the ability to safely let AI agents optimize the code, relying on the formal proof to guarantee correctness.