Prime Intellect launches Prime Mega MoE, a trainable fused MoE kernel
Prime Intellect unveiled Prime Mega MoE, a trainable fused MoE kernel that speeds forward passes 3.11× and cuts Qwen3 training time by up to 34%.
Try in Grafy Codemega moekerneltrainingai