Cerebras has released MiniMax-M2-REAP-162B-A10B, a compressed Sparse Mixture-of-Experts (SMoE) Causal Language Model derived from MiniMax-M2, using the new Router weighted Expert Activation Pruning ...