The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction
Edge0 predicts next-layer MoE routing one token ahead to stream experts from SSD, serving 35B-class MoEs at 20 tok/s within 3 GiB active memory on a 24 GB machine via recovery LoRA adapters.
Published Sep 16, 2026▲ 25 on Hugging FaceCode ★ 3,293arXiv ↗
Only vote on papers you've read. Sign in with GitHub to vote.
Edge0 achieves striking 35B MoE throughput on consumer SSDs via predictive routing and recovery LoRA, but hides critical evaluation gaps behind unnamed benchmarks, missing long-context routing stress tests, and unreported misprediction amplification.
Abstract
Mixture-of-experts (MoE) inference on consumer hardware is bounded by weight memory: a 35B-class model is 19.5GB at 4-bit, and sparsity shrinks the compute per token, not the bytes that must be held. Naive offloading to SSD does not help on its own, because layer N+1's experts must be chosen before layer N's output exists, so the reads cannot start early enough to hide behind compute. We present Edge0, a streaming MoE inference engine that closes the gap with a prerouter: a per-layer head predicts the next layer's routing one token ahead, and the prediction is consumed as the routing itself, so the staged expert set equals the routed set and nothing is dropped. An unmerged recovery LoRA, trained on the student path, pays back the quality lost to int4 quantization and routing replacement. On a single 24GB machine, Edge0 serves a 35B MoE at 20tok/s inside 3GiB of peak active memory, within a few points of its fp16 teacher on average across five public benchmarks. An 8B tier runs on the same framework, and the framework, checkpoints, and adapters are open source.