MawForge: Memory-Bounded Expert Materialization for Local Mixture-of-Experts Inference

Researchers introduced MawForge, a system for running sparse Mixture-of-Experts language models on memory-constrained local machines by storing the full model on disk and materializing expert tensors into a bounded cache on demand. The approach is effective as a measurement substrate but not as a cache-maximization policy, with performance depending on factors like expert reuse and memory pressure.

arXiv:2607.09686v1 Announce Type: new Abstract: Sparse Mixture-of-Experts MoE language models separate total parameter count from per-token active computation, but local inference systems often still require the full model, key-value cache, runtime buffers, and operatingsystem headroom to fit in fast memory. MawForge tests a different systems hypothesis: local MoE serving can be made practical on constrained unified-memory machines by storing the full model on disk, keeping common tensors resident, and materializing routed expert tensors into a bounded execution cache on demand. The central finding is that MawForge is effective as a bounded execution mechanism and measurement substrate for local MoE inference, but not as a cache-maximization policy. Performance depends on balancing expert reuse against resident footprint, KV-cache size, quantization, route locality, and macOS memory pressure.