AI 中文总结
本文提出 Routide 运行时，在 iPhone 上以闪存存储专家权重，通过缓存策略优化实现 MoE 推理，验证了有界可行性并揭示了容量悬崖源于策略与工作负载交互。
AI 中文摘要
稀疏激活减少了混合专家模型的计算量，但并未消除存储所有专家的需求。我们提出了 Routide，一个 Swift/MLX 运行时，它在保持专家权重位于 iPhone 存储中且内存中仅保留字节预算子集的同时，执行一个固定的公开 Qwen3.6-35B-A3B 量化检查点的文本路径。我们描述了缓存策略敏感性、数值比较边界和测量限制。在五个记录的 128-token 工作负载中，固定路由重放使用 512 MiB LRU 缓存时需求命中率为 0.00%，使用相同预算的种子随机驱逐时为 18.80%，使用 576 MiB LRU 时为 38.58%。因此，明显的容量悬崖是策略/工作负载交互的结果，而非普遍的内存需求。同运行时的 Mac 对照在驱逐和异步预取过程中保持了生成的序列，包括 2,560 次精确 token 比较和 10,334 次推测性加载。相比之下，完整的驻留 Python 与记录的手机序列在所有五个测试案例上都不一致，排除了普遍数值等价性的主张。两个分别限定的 iOS 27 内存协议观察到采样进程占用峰值在短提示上为 1.87-2.32 GiB，在一个较长提示上为 2.39-2.73 GiB。我们保留了一个热停止事件、负时序比较和单个合格的整机功耗估计。这些结果确立了有界可行性，并识别了部署声明不得隐藏的限制。
英文摘要
Sparse activation reduces mixture-of-experts computation without eliminating the need to store all experts. We present Routide, a Swift/MLX runtime that executes the text path of a pinned public Qwen3.6-35B-A3B quantized checkpoint while keeping expert weights in iPhone storage and a byte-budgeted subset in memory. We characterize cache-policy sensitivity, numerical comparison boundaries, and measurement limits. Across five recorded 128-token workloads, fixed-route replay gives 0.00% demand hits with a 512 MiB LRU cache, 18.80% with seeded random eviction at the same budget, and 38.58% with 576 MiB LRU. The apparent capacity cliff is therefore a policy/workload interaction, not a universal memory requirement. Same-runtime Mac controls preserve generated sequences across eviction and asynchronous prefetch, including 2,560 exact token comparisons and 10,334 speculative loads. In contrast, complete resident-Python versus recorded-phone sequences disagree on all five tested cases, precluding a general numerical equivalence claim. Two separately scoped iOS 27 memory protocols observe sampled process-footprint peaks of 1.87-2.32 GiB on short prompts and 2.39-2.73 GiB on one longer prompt. We retain a thermal stopping event, negative timing comparisons, and a single qualified whole-device power estimate. These results establish bounded feasibility and identify limitations that a deployment claim must not hide.
Comments13 pages, 2 figures. Code: https://github.com/MusaShams/Routide
