AIMarkTechPost1 day ago
DeepSeek DSpark Speeds V4 Inference 60-85%
7 min read
DeepSeek released DSpark, a speculative decoding framework that accelerates DeepSeek-V4 per-user generation by 60 to 85 percent over MTP-1. Open-source checkpoints and training code are available. DeepSpec, an MIT-licensed codebase for training drafters, was also open-sourced.
Level
Hype check
Tap to vote and see what everyone thinks.
