MojoLlama is a high-throughput inference engine for CPU, built on Modular MAX. GGUF native, MoE-optimized, with support for 50+ architectures — from Llama to Gemma 4 to hybrid SSM models. GPU acceleration via MAX engine for supported models.