Moe inference optimizations: 15% lower expert load by request reordering

(blog.doubleword.ai) by mezark | May 20, 2026 | 0 comments on HN