← 研踪 PaperTrailarXiv:2610.22454 · v3
Where Prefill Expert Reduction Hurts MoE Models: A Few Prompt Tokens Carry Most of the Short-Answer F1 Loss
首次提交 2026-10-16 00:26
版本更新 2026-11-04 08:23
雷达观测 2026-11-05 10:08
摘要
Running prompt tokens through fewer routed experts during prefill makes mixture-of-experts (MoE) serving cheaper, but what it damages has not been measured. We locate the damage with token-, layer- and attention-path interventions in four MoE families on short-answer question answering. At half the experts the F1 loss is mostly answer form: the model still finds the answer but no longer answers briefly. The loss is made at a few tokens: the two chat-template tokens before the answer in Qwen3-30B-A3B, and the first 8-32 tokens, where the instruction sits, in DeepSeek-V2-Lite. Keeping them at full experts removes most of the F1 loss and about half and a fifth of the increase in answer NLL. Reduction changes the first-token sink statistics by less than 4%, and restoring super experts or massive activations recovers at most a quarter of the loss. We then test placement on real prompts with about 4% of tokens at full experts. Putting them on the task interface keeps F1 within one point of the unmodified model, while the same budget scattered in the documents is about 3 points worse in both prespecified primary tests of this comparison. Time to first token falls about 14% on 11K-token prompts.
作者备注：8 pages main text, 26 pages with appendices; supplementary material in ancillary files
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-04 08:23 · 最新观测版本
Where Prefill Expert Reduction Hurts MoE Models: A Few Prompt Tokens Carry Most of the Short-Answer F1 Loss
Running prompt tokens through fewer routed experts during prefill makes mixture-of-experts (MoE) serving cheaper, but what it damages has not been measured. We locate the damage with token-, layer- and attention-path interventions in four MoE families on short-answer question answering. At half the experts the F1 loss is mostly answer form: the model still finds the answer but no longer answers briefly. The loss is made at a few tokens: the two chat-template tokens before the answer in Qwen3-30B-A3B, and the first 8-32 tokens, where the instruction sits, in DeepSeek-V2-Lite. Keeping them at full experts removes most of the F1 loss and about half and a fifth of the increase in answer NLL. Reduction changes the first-token sink statistics by less than 4%, and restoring super experts or massive activations recovers at most a quarter of the loss. We then test placement on real prompts with about 4% of tokens at full experts. Putting them on the task interface keeps F1 within one point of the unmodified model, while the same budget scattered in the documents is about 3 points worse in both prespecified primary tests of this comparison. Time to first token falls about 14% on 11K-token prompts.
查看此版本解读 →查看此版本原文 ↗