Faster AI, Lower Costs: DSpark Eases Inference Bottlenecks and Chip Strain, Says DeepSeek

Faster AI, lower costs: DSpark eases bottlenecks and chip strain, says DeepSeek

Start-up unveils speculative decoding framework that speeds up inference by up to 85 per cent amid China’s push to overcome US AI curbs.

Faster AI, lower costs: DSpark eases inference bottlenecks and chip strain, says DeepSeek

Chinese artificial intelligence start-up DeepSeek has rolled out a major upgrade to its flagship V4 model aimed at sharply accelerating AI response generation, as competition among Chinese developers increasingly shifts to reducing serving costs and enhancing user experience. DeepSeek, by adopting…

Peking University and DeepSeek Open-Source DSpark, Delivering Major Leap in LLM Inference Efficiency

Peking University and DeepSeek have jointly released and open-sourced DSpark, a speculative decoding framework that delivers a major leap in large language model inference efficiency, achieving 60-85% faster text generation under real-world server loads and up to 661% throughput improvement under strict latency constraints. Current large language models rely on autoregressive generation, where each token requires a full forward pass, creating si…

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 probl…