Moe inference optimizations: 15% lower expert load by request reordering(blog.doubleword.ai)
blog.doubleword.ai
Moe inference optimizations: 15% lower expert load by request reordering
https://blog.doubleword.ai/moe-expert-coactivations
0 comments
—
https://blog.doubleword.ai/moe-expert-coactivations
—