jaxytee·4 gün önce·discussRunning large MOE models won't saturate the EPYC memory.You'll never see that 200 Gb/s running inference.
You'll never see that 200 Gb/s running inference.