Kranium2002/cacheshrink

Python library that converts HuggingFace transformer models to use Multi-Head Latent Attention (MLA) with Riemannian optimization on Stiefel manifolds.

5 best points · 2 comments · 1 Hacker News post · 8 GitHub stars · View on GitHub

Hacker News discussions

• KV Cache Compression (5 points, 2 comments, by Kranium2002, 2026-02-18)