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

3 pointsposted 11 hours ago
by mezark
(blog.doubleword.ai)

No comments yet