DSpark breaks through the ceiling of speculative decoding in DeepSeek-V4
bycloud · youtube · 2026-07-21
DeepSeek-V4 compute optimization, broken down in DSpark
The video walks through DSpark, a paper on DeepSeek-V4’s inference-side optimization strategy. The core claim is that the lab pushed speculative decoding much further than usual, combining theory with practical serving considerations.
The presenter frames it as a rare example of a lab sharing aggressive compute min-maxing in a way that is both technically interesting and operationally useful for deployment.
More from Infra
- Spomin: live KV cache compaction squeezes 500k tokens of context into 180k resident — wgaca2 · 2026-09-11
- PiPNN nearest-neighbor search wins three awards, up to 78x faster index building — khademinori · 2026-09-11
- M.2-Oculink eGPU Link Silently Downgrades to PCIe Gen1 — Here's How to Check — El_90 · 2026-09-11
- DeepSeek launches V4.1-Flash with 1M-token context and 4x smaller KV-cache — matlabulous · 2026-09-11
- What Can You Still Run on 8GB VRAM? User Asks for Small Models With Tool Use — riceinmybelly · 2026-09-11
- Spain's hourly 80% renewable matching rules clash as France fast-tracks 700MW sites, UK cuts grid queues — eherrerosj · 2026-09-11