DeepSeek Releases DSpark, a Speculative Decoding Framework That Accelerates DeepSeek-V4 Per-User Generation 60–85% Over MTP-1

DeepSeek released DSpark, a speculative decoding framework, with open-source checkpoints and training code. It is a serving optimization, not a new model. The checkpoints DeepSeek-V4-Pro-DSpark and DeepSeek-V4-Flash-DSpark reuse the existing V4 weights, with a draft module attached.

The DeepSeek research team also open-sourced DeepSpec, an MIT-licensed codebase for training and evaluating speculative decoding drafters. The work targets one problem: faster large-model inference in busy production serving.

TL;DR

DSpark pairs a parallel draft backbone with a tiny sequential head to cut suffix decay.

A confidence head and load-aware scheduler verify more tokens when GPUs are idle, fewer when busy.