Scalable Analyses · @scalable
8 followers · 126 posts · Server fosstodon.org

Meta designed the “Meta Training and Inference Accelerator” which meets demands of deep learning recommender models at inference time. The accelerator is powered by a grid of processing elements where each element contains two RISC-V cores (one of these has a vector extension): nextplatform.com/2023/05/18/me

#meta #mtia #DLRM #inference #ai #hpc

Last updated 1 year ago