howtospark
Training

DSpark

Accelerate decodingSpeculative decodingReference

DeepSeek's drafter: a parallel backbone proposes a whole block at once, a lightweight serial head glances one token back to stop the tail drifting, and a confidence head schedules how far to verify — so it doesn't pay to draft-and-discard past the acceptance knee.

On the Spark

Runs on 2× DGX Spark via vLLM --speculative-config '{"method":"dspark",...}'; the confidence-scheduled block is the right shape for GB10's bandwidth-bound decode, where a fixed large k wastes the shared memory bus. See models/ DSpark bring-up log.

Objective
Accelerate decoding
Relieves
Bandwidth
Format
Semi-autoregressive parallel drafter + confidence-scheduled verify
Granularity
Block draft (parallel backbone + serial Markov head)
Lifecycle
QAT / QAD · needs gradients
Calibration
Small calibration set
Compression
~1.6–1.9× decode throughput (semi-autoregressive block draft)
Quality
100% — verification keeps only tokens the target model would emit
Hardware
Trained DSpark drafter (baked into the repo or a companion speculator repo); confidence-scheduled verify kernel
Runtimes
vLLM
In the wild
DeepSeek DSpark + DeepSpec

Open-sourced DSpark checkpoints plus DeepSpec, the training repo — 57–85% faster inference, lossless, on V4 / Qwen / GLM / Gemma.

RedHatAI/GLM-5.2-speculator.dspark

Companion DSpark drafter repo for GLM-5.2 — the 'ships a speculator' pattern the optimizer detects.

NVIDIA NeMo AutoModel — Train a DSpark Drafter

End-to-end recipe to train and export a DSpark drafter for a target model.

llama.cpp: add DSpark speculative decoding (PR #25173)

In-flight llama.cpp support — DSpark reaching the local-inference stack beyond vLLM.

A worked Spark recipe for this method hasn't been written yet — it lives here as a reference point in the ontology.