Understanding Irregular MoE Execution on Modern GPUs理解现代 GPU 上的不规则 MoE 执行

Analyzing routing, token distribution, and execution behavior without reducing irregular workloads to an average case.分析路由、Token 分布与执行行为,而不把不规则负载简化成平均情形。

Working outline研究提纲 · Hypotheses and measurement plans; verified results are still being attached.包含假设与测量计划,经验证的结果仍在补充中。

Why irregularity matters

Mixture-of-experts execution inherits variability from routing. A useful analysis preserves the distribution rather than reporting only its mean.

Measurement plan

TODO: record the verified workload, hardware, runtime revision, and profiler captures.

为何不规则性重要

混合专家模型的执行会继承路由带来的变化。有效的分析应保留完整分布,而不是只报告平均值。

测量计划

TODO:记录经过验证的负载、硬件、运行时版本和性能分析截图。

vLLM · Next in this project接着读GPU Memory Behavior in Long-Context Inference长上下文推理中的 GPU 内存行为 →Working outline研究提纲