Liked it? Make your own.
How Cerebras Serves a 3-Trillion-Parameter Model on 59 Wafers
Size it, split it, chain it, fill it: how Kimi K3's 1.5 TB of weights spread across 59 wafers to serve about 190 requests at once.
Liked it? Make your own.
Size it, split it, chain it, fill it: how Kimi K3's 1.5 TB of weights spread across 59 wafers to serve about 190 requests at once.