DeepSeek's New Paper Proposes DualPath Inference System, Doubling Agent Workload Throughput

Amid the industry's anticipation for the next-generation flagship model DeepSeek V4, the DeepSeek team has quietly released a new academic paper. The paper introduces an innovative inference system called DualPath, specifically designed to optimize the inference performance of large models (LLMs) under agent workloads. By introducing a 'DualPath KV-Cache (similar to memory cache)' mechanism, it redistributes storage network loads, achieving up to a 1.87x increase in offline inference throughput and an average 1.96x increase in the number of agents running per second in online services. The introduction section of the paper mentions that large models are rapidly evolving from single-turn conversational bots and standalone inference models into agent systems—capable of autonomous planning, tool invocation, and solving real-world tasks through multi-turn interactions. This paradigm shift in application is driving significant changes in large model inference workloads: transitioning from traditional human-large model interaction to human-large model-environment interaction, with interaction rounds reaching dozens or even hundreds.