{"slug": "dpa-moe-a-runtime-precision-control-framework-for-moe-inference", "title": "DPA-MoE: A Runtime Precision Control Framework for MoE Inference", "summary": "Researchers introduced DPA-MoE, a runtime precision control framework that encodes Mixture-of-Experts feed-forward weights as prefix-decodable chunks using SharedScale AnyBCQ and allocates chunks at inference time based on live router gate scores, letting a single encoded checkpoint run at multiple effective weight budgets without retraining or re-encoding. Experiments on Qwen3-30B-A3B and DeepSeekMoE-16B showed DPA-MoE improved KL divergence and downstream accuracy over uniform chunk allocation at matched budgets, with modest steady-state throughput overhead in the prototype.", "body_md": "arXiv:2610.21378v1 Announce Type: new \nAbstract: Mixture-of-Experts (MoE) models reduce computation by routing each token to a small set of experts, but each selected expert is still used at a fixed precision. We present DPA-MoE, a runtime precision control framework that makes routed experts any-precision. DPA-MoE encodes expert feed-forward weights as prefix-decodable chunks using SharedScale AnyBCQ, profiles per-expert reconstruction residuals offline, and allocates chunks at inference time using live router gate scores. This post-routing allocation lets a single encoded MoE checkpoint operate at multiple effective weight budgets without retraining or re-encoding. Experiments on Qwen3-30B-A3B and DeepSeekMoE-16B show that DPA-MoE improves KL divergence and downstream accuracy over uniform chunk allocation at matched budgets, with modest steady-state throughput overhead in our prototype.", "url": "https://wpnews.pro/news/dpa-moe-a-runtime-precision-control-framework-for-moe-inference", "canonical_source": "https://www.machinebrief.com/news/dpa-moe-a-runtime-precision-control-framework-for-moe-infere-x3nx", "published_at": "2026-10-16 04:00:00+00:00", "updated_at": "2026-10-16 05:36:43.314547+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-research", "ai-infrastructure"], "entities": ["DPA-MoE", "SharedScale AnyBCQ", "Qwen3-30B-A3B", "DeepSeekMoE-16B", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/dpa-moe-a-runtime-precision-control-framework-for-moe-inference", "markdown": "https://wpnews.pro/news/dpa-moe-a-runtime-precision-control-framework-for-moe-inference.md", "text": "https://wpnews.pro/news/dpa-moe-a-runtime-precision-control-framework-for-moe-inference.txt", "jsonld": "https://wpnews.pro/news/dpa-moe-a-runtime-precision-control-framework-for-moe-inference.jsonld"}}