Tag: Wafer-scale

Cerebras says splitting inference in two yields five times the throughput

By running prompt processing and token generation on separate hardware pools, Cerebras says it served five times more…