NohWai-Software/RayTention

Raytention is a new attention mechanism that solves the bloated k-v cache VRAM problem. Raytention utilizes signals built from the context to provide the model with the attention it needs at a much lower VRAM cost.

2 best points · 0 comments · 2 Hacker News posts · 4 GitHub stars · View on GitHub

Hacker News discussions

• RayTention – Self-Attention via Geometric Signal Extraction (2 points, 0 comments, by NohWai, 2026-07-01)
• Show HN: Raytention Updated (1 points, 0 comments, by NohWai, 2026-07-19)