▲ 1 Measured attention kernels vs. AMD's own on MI350X, with every loss published (github.com) by wonka-ai | Nov 19, 2026 | 0 comments on HN Visit Link