Files
myWiki/concepts/MTP.md

31 lines
1.3 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

---
title: "MTP (Multi-Token Prediction)"
created: 2026-06-28
updated: 2026-06-28
type: concept
tags: [speculative-decoding, autoregressive-drafting, deepseek, production]
sources: [DSpark]
---
# Multi-Token Prediction (MTP)
MTPMulti-Token Prediction是 DeepSeek 系列模型DeepSeek-V2, V3, V4内置的[[autoregressive-drafting|自回归草稿Autoregressive Drafting]]机制,作为生产环境中的投机解码草稿器。
## 变体
- **MTP-1**:单 token 草稿模式DSpark 论文中的生产基线。历史上被维持是因为静态多 token 草稿器MTP-3/5在高并发下会因过度验证开销严格降低总吞吐量
- **MTP-3/5**:多 token 草稿模式
## 在 DSpark 部署中的角色
DSpark 在 DeepSeek-V4 发布两周后取代了 MTP-1 成为生产系统。DSpark-5最大草稿长度 $\gamma=5$)相比 MTP-1
- V4-Flash在匹配吞吐量下每用户生成速度提升 60%-85%
- V4-Pro提升 57%-78%
关键突破DSpark 通过[[confidence-scheduled-verification|置信度调度验证Confidence-Scheduled Verification]]安全解锁了大草稿块的性能潜力,解决了 MTP-3/5 因固定长度验证导致的吞吐量退化问题。
## 参考
- [[DSpark]]
- [[autoregressive-drafting|自回归草稿Autoregressive Drafting]]
- [[Eagle3]]