Open-Sourced DSpark Speculator: Boosts Inkling Decode Throughput by 1.89X
BanghuaZ · x · 2026-07-23
A developer has trained a new DSpark speculator for the Inkling NVFP4 model, built end-to-end with SpecForge on a live SGLang target engine, significantly improving inference efficiency.
Performance metrics include:
- 1.89x decode throughput improvement over non-speculative decoding at bs=64 (8xB200, TP8).
- 7-14% faster than Inkling's built-in MTP at the same batch size.
- Mean accept length of 3.66, reaching up to 4.76 on the GSM8K dataset.
Under the hood, it utilizes a 5-layer Qwen3-style DFlash parallel-draft backbone, a Rank-256 Markov logit-bias head, a per-position confidence head for acceptance prediction, and was distilled online from 400K Inkling regenerations.
Related event: Open-Sourced DSpark Speculator Boosts Inkling Throughput by 1.89x(3 posts)→
More from Infra
- 12 KV Cache Reduction Techniques Every AI Engineer Should Understand, Explained — blaizedsouza · 2026-09-11
- The shadow GPU capacity market is formalizing, with Meta selling excess compute to outside buyers — DavidLinthicum · 2026-09-11
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11