1 paper
Itay Lamprecht, Asaf Karnieli, Yair Hanani +2
Training and inference of Large Language Models (LLMs) with tensor-parallelism requires substantial communication to synchronize activations. Our findings suggest that with a few m…