EVENT DOSSIER
NVIDIA released NeMo Switchyard and published test data on intelligent routing, showing that only 7% of tasks required advanced models, with costs reduced by 7…
NVIDIA has introduced NeMo Switchyard, designed to schedule AI Agent workloads across models through intelligent routing. In tests involving 145 agent tasks, it was found that only 7% of dialogue rounds required advanced models. Compared to sending all requests to the most powerful model, NeMo Switchyard improves accuracy by six points while reducing costs by 74%. This tool allows users to dynamically allocate different models based on task requirements, such as using classification models for specific steps, reasoning models for the next step, and small models for routine follow-up tasks, thereby optimizing model selection and resource allocation in the AI Agent construction process.
企业 · 官方发布across 1 days
NVIDIA released NeMo Switchyard and published test data on intelligent routing, showing that only 7% of tasks required advanced models, with costs reduced by 7…
NVIDIA introduced NeMo Switchyard for intelligent routing. In 145 task tests, only 7% required advanced models, with six points of accuracy improvement and 74% cost reduction.
2 reportsNVIDIA 推出 NeMo Switchyard，旨在跨模型调度 AI Agent 工作负载。该工具允许用户根据任务需求动态分配不同模型，以利用各模型的特定优势、弱点及成本特征。例如，单一代理任务可能需分类模型处理某一步骤、推理模型处理下一步，以及小模型处理常规跟进任务。若将所有请求发送至最大模型将导致效率低下或成本增加。NeMo Switchyard 通过智能路由解决此问题，优化 AI Agent 构建流程中的模型选择与资源分配。
NVIDIA 在 145 个智能体任务中测试 NeMo Switchyard，发现仅 7% 的对话回合需要前沿模型，且该路由方案在提升六点准确度的同时降低了 74% 的成本。