适用于网络运营中心的生成式 AI

利用生成式 AI 安全地规划、构建和运营电信网络。

工作负载

生成式 AI

行业

电信

业务目标

风险缓解

产品

NVIDIA AI Enterprise
NVIDIA NIM 微服务
NVIDIA NeMo

生成式 AI 加速网络配置、部署和运营

2024 年，电信公司的资本支出 (CapEx) 估计接近 2950 亿美元，运营支出 (OpEx) 超过 1 万亿美元，其中包括基于人工的网络规划和维护流程支出。在电信网络中，配置和优化涉及管理大量相互依赖的参数，这些参数直接影响数百万客户和最终用户的网络性能、用户体验和频谱效率。电信网络工程师需要根据一天中的时间、用户行为、移动性、干扰和服务类型不断调整这些设置。

生成式 AI 为大型电信模型 (LTM) 和 AI 智能体提供支持，可在网络运营中实现新一代 AI，支持电信公司优化运营支出 (OpEx)、高效利用资本支出 (CapEx)，并发掘新的盈利机会。NVIDIA 开发了一种代理式 AI 解决方案，通过观察实时网络 KPI、做出数据驱动型决策和自动调整参数，将自主性引入这一动态环境。

与传统的基于规则的系统不同，AI 智能体可以感知、通过复杂的权衡进行推理、从反馈回路中学习，并根据需要添加人类在环反馈来适应新的条件。它还可以在多个层和多个供应商之间编排更改，从而实现负载平衡、单元间干扰协调或在负载较轻的区域节能等协调行动。这种级别的自主控制不仅提高了效率和服务质量 (QoS)，还降低了运营复杂性，并缩短了在密集、高需求环境中解决问题的时间。

 

快速链接

阅读博客：调用 LLM：全新的 NVIDIA AI Blueprint 推动电信网络配置自动化
阅读博客：电信行业领先企业启用代理式 AI 优化网络运维
阅读报告：NVIDIA“电信行业 AI 现状”
阅读案例研究：AT&T 借助 NVIDIA 提升 AI 智能体的准确性、效率和性能

Boosting Network Performance and Efficiency With Accelerated Computing

Global telecommunications companies are exploring how to cost-effectively deliver new AI applications to the edge over 5G and upcoming 6G networks. With NVIDIA accelerated computing and AI, telcos, cloud service providers (CSPs), and enterprises can build high-performance cloud-native networks—both fixed and wireless—with improved energy efficiency and security. 

The NVIDIA AI Foundry for Generative AI

The NVIDIA AI foundry—which includes NVIDIA AI Foundation models, the NVIDIA NeMo™ framework and tools, and NVIDIA DGX™ Cloud—gives enterprises an end-to-end solution for developing custom generative AI. 

Amdocs, a leading software and services provider, plans to build custom large language models (LLMs) for the $1.7 trillion global telecommunications industry using the NVIDIA AI foundry service on Microsoft Azure. In network operations, Amdocs and NVIDIA are exploring ways to generate solutions that address configuration, coverage, and performance issues as they arise, including:  

  • Building a generative AI assistant to answer questions around network planning
  • Providing insights and prioritization for network outages and performance degradations
  • Optimizing operations by using generative AI to monitor, predict, and resolve network issues, manage resources in real time​, monitor network diagnostics, analyze service and user impact, prioritize impact-based recommendations, and execute orchestration activation

 

ServiceNow is integrating generative AI capabilities into their Now Platform and enriching all workflows with Now Assist, their generative AI assistant. ServiceNow leverages NeMo and NVIDIA Triton™ Inference Server (both part of NVIDIA AI Enterprise), NVIDIA AI Foundation models, and DGX systems to build, customize, and deploy generative AI models for telecom customers. These include use cases in network operations:

  • Automated service assurance: Distill and act on volumes of complex technical data generated from network incidents​ and summarized by generative AI.
  • Streamlined service delivery​: Dynamically create order tasks with generative AI to reduce human errors, ensure accurate service delivery, and improve customer satisfaction and loyalty.
  • Optimized network design: Manage diverse network services, local configurations, and rulings to improve network design.

 

NeMo provides an end-to-end solution—including enterprise-grade support, security, and stability—across the LLM pipeline, from data processing to training to inference of generative AI models. It allows telcos to quickly train, customize, and deploy LLMs at scale, reducing time to solution while increasing return on investment.

The NVIDIA AI foundry includes NVIDIA AI Foundation models, the NeMo framework and tools, and NVIDIA DGX™ Cloud , giving enterprises an end-to-end solution for creating custom generative AI models.

Once generative AI models are built, fine-tuned, and trained, NeMo enables seamless deployment through optimized inference on virtually any data center or cloud. NeMo Retriever, a collection of generative AI microservices, provides world-class information retrieval with the lowest latency, highest throughput, and maximum data privacy, enabling organizations to generate insights in real time. NeMo Retriever enhances generative AI applications with enterprise-grade retrieval-augmented generation (RAG), which can be connected to business data wherever it resides.

NVIDIA DGX Cloud is an AI-training-as-a-service platform, offering a serverless experience for enterprise developers that’s optimized for generative AI. Enterprises can experience performance-optimized, enterprise-grade NVIDIA AI Foundation models directly from a browser and customize them using proprietary data with NeMo on DGX Cloud.

NVIDIA AI Enterprise for Accelerated Data Science and Logistics Optimization

The NVIDIA AI Enterprise software suite enables quicker time to results for AI and machine learning initiatives, while improving cost-effectiveness. Using analytics and machine learning, telecom operators can maximize the number of completed jobs per field technician​, dispatch the right personnel for each job, dynamically optimize routing based on real-time weather conditions​, scale to thousands of locations​, and save billions of dollars in maintenance.

AT&T is transforming their operations and enhancing sustainability by using NVIDIA-powered AI for processing data, optimizing fleet routing, and building digital avatars for employee support and training. AT&T first adopted the NVIDIA RAPIDS™ Accelerator for Apache Spark to capitalize on energy-efficient GPUs across their AI and data science pipelines. Of the data and AI pipelines targeted with Spark RAPIDS, AT&T saves about half of their cloud computing spend and sees faster performance, while reducing their carbon footprint.

AT&T, which operates one of the largest field dispatch teams, is currently testing NVIDIA® cuOpt™ software to to handle more complex technician routing and optimization challenges. In early trials, cuOpt delivered solutions in 10 seconds, while the same computation on x86 CPUs took 1,000 seconds. The results yielded a 90 percent reduction in cloud costs and allowed technicians to complete more service calls each day.

Quantiphi, an innovative AI-first digital engineering company, is working with leading telcos to build custom LLMs to support field technicians​. Through LLM-powered virtual assistants acting as copilots, Quantiphi is helping field technicians resolve network-related issues and manage service tickets raised by end customers.

“Ask AT&T was originally built on OpenAI’s ChatGPT functionality. But Ask AT&T is also interoperable with other LLMs, including Meta’s LLaMA 2 and the open-source Falcon transformers. We’re working closely with NVIDIA to build and customize LLMs. Different LLMs are suited for different applications and have different cost structures, and we’re building that flexibility and efficiency in from the ground floor.”

Andy Markus, Chief Data Officer, AT&T

用于网络运营的生成式 AI 入门

NVIDIA AI Blueprint 通过提供可供开发者创建自己的 AI 智能体的工作流，实现了可扩展的自动化。借助这些工具，开发者可以构建和部署自定义 AI 智能体，这些智能体可以进行推理、规划并采取行动，快速分析大量数据、总结并提炼实时见解。

适用于电信网络配置的 NVIDIA AI Blueprint 为跨多个域的网络运营提供经过验证的基础模组。此 AI Blueprint 使开发者、网络工程师、电信公司和供应商能够使用代理式 LLM 驱动的框架自动配置无线接入网 (RAN) 参数。

自主网络为更好地管理运营支出（OpEx）提供了机会。适用于电信网络配置的 AI Blueprint 通过模块化的 AI 架构以及实现一致、可扩展部署所需的自动化工作流程，为这一目标的达成提供了助力。此 AI Blueprint 由生成式 AI 提供支持，使网络工程师能够增添自适应智能，从而预测问题、优化性能并实现决策自动化。

BubbleRAN 和 Telenor 采用 NVIDIA AI Blueprint 进行电信网络配置

用于电信网络配置的 AI Blueprint 由云原生基础设施上的 BubbleRAN 软件提供支持，可用于大规模构建自主网络及其多智能体 RAN 智能平台。

Telenor Group 为全球超过 2 亿客户提供服务，计划为电信网络配置部署 AI Blueprint，以解决网络安装期间的配置挑战并增强 QoS。

实施详情

此代理式 LLM 驱动框架利用 Llama 3.1-70B-Instruct 作为基础 AI 模型，这是因为该模型在自然语言理解、推理以及工具调用方面表现出色。

客户可以通过以下方式灵活部署此 Blueprint：

  • NVIDIA 托管的 NIM™ 微服务 API 端点，请访问 build.nvidia.com
  • 满足隐私和延迟要求的本地 NIM 微服务

终端用户通过基于 Streamlit 的用户界面 (UI) 进行查询或启动网络操作。这些请求由 LangGraph 智能体框架进行处理，该框架负责协调各个专门的 LLM 智能体任务。

LLM 智能体配备了专用工具，能够针对实时和历史 KPI 数据生成并执行 SQL 查询，计算所收集数据的加权平均收益，应用配置变更，并处理 BubbleRAN 环境相关事务。

我们采用 prompt-tuning 技术，注入有关 BubbleRAN 网络架构的上下文知识，包括网络架构的设置细节、各项 KPI 指标之间的相互依赖关系以及用于平衡权衡以优化加权平均收益的逻辑。

由 LangGraph 提供支持的代理式框架统筹管理三个具备不同职责的专门化智能体，它们协同工作，共同完成监控、配置和验证的完整闭环流程。用户使用选定的参数完成网络初始化后，可以选择使用监控代理进行监控会话，也可以直接查询配置代理以了解参数对网络的影响以及当前网络状态。

快速链接

开始使用 AI Blueprint
Architecture diagram of NVIDIA AI Blueprint for telco network configuration.

以下是每个智能体及其功能的细分：

1. 监控智能体
该智能体在实时 BubbleRAN KPI 数据库中，持续追踪用户自定义时间间隔（默认：10 秒）内预设参数的平均加权收益。当检测到因某一特定参数的加权平均收益降低而导致性能下降时，它会将问题上报给用户，以征求用户对下一步操作的授权。

2. 配置智能体
配置智能体可由监控智能体移交任务而激活，也可在用户直接查询参数优化或网络健康状况时激活。它会分析历史数据，然后依据分析得出的趋势以及参数间相互依赖关系和权衡取舍方面的领域特定知识进行推理。基于分析结果，它会向用户提出改进后的参数值建议，并等待用户确认。

3. 验证智能体
确认参数调整后，验证智能体就会使用新的参数配置重新启动网络。它会在用户可配置的验证期内对更新后的参数进行评估，并计算得出的平均加权收益。如果实时平均加权收益进一步恶化，它会自动回滚到之前的稳定配置。否则，它会确认操作成功，并在用户界面上更新新设置。

总之，我们的框架通过智能体循环实现了持续、智能的网络优化，其中专门的 LLM 智能体协同工作，实时监控、分析和验证参数变化。这些智能体配备了分析实时和历史 KPI 数据的工具，并具备网络参数和权衡取舍方面的领域特定知识，可提供数据支持的建议和可解释推理。这种闭环设计确保了网络性能既保持自主性又具备用户可控性，使用户能够在保持对每个决策点控制的同时，维持网络的最优性能。

有关更多技术细节，请浏览 Blueprint。

NVIDIA NIM

NVIDIA NIM, part of NVIDIA AI Enterprise, is an easy-to-use runtime designed to accelerate the deployment of generative AI across your enterprise. This versatile microservice supports open community models and NVIDIA AI Foundation models from the NVIDIA API catalog, as well as custom AI models. NIM builds on NVIDIA Triton™ Inference Server, a powerful and scalable open source platform for deploying AI models, and is optimized for large language model (LLM) inference on NVIDIA GPUs with NVIDIA TensorRT-LLM. NIM is engineered to facilitate seamless AI inferencing with the highest throughput and lowest latency, while preserving the accuracy of predictions. You can now deploy AI applications anywhere with confidence, whether on-premises or in the cloud.

NVIDIA NeMo Retriever

NeMo Retriever is a collection of CUDA-X microservices enabling semantic search of enterprise data to deliver highly accurate responses using retrieval augmentation. Developers can use these GPU-accelerated microservices for specific tasks including ingesting, encoding, and storage large volumes of data, interacting with existing relational databases, and searching for relevant pieces of information to answer business questions.

生成式 AI 可以分析来自设备传感器的大量数据，以预测潜在的故障或问题。这有助于技术人员在问题发生之前进行预判，从而及时进行维护，减少停机时间。

基于生成式 AI 的分析根据实时数据为技术人员提供切实可行的见解和建议。这使他们能够在维修、升级和网络优化方面做出明智的决策。

生成式 AI 可以自动执行重复的常规任务，例如生成工单、安排预约和创建文档。这使技术人员能够将更多精力投入到复杂问题和客户服务上。

快速链接

探索适用于电信运营的 AI

利用生成式 AI 优化网络运营

通过利用 NVIDIA AI，电信公司可以减少网络宕机时间，提高现场技术人员的工作效率，并为客户提供更好的服务质量。联系我们的专家团队或探索其他资源。

联系销售人员
观看视频

Resources

Generative AI in Practice: Examples of Successful Enterprise Deployments

Learn how telcos built mission-critical LLMs, powered by NVIDIA DGX systems and the NeMo framework, to simplify their business, increase customer satisfaction, and achieve the fastest and highest return.

Read Ebook

Part 1: A Beginner's Guide to Large Language Models

Get an introduction to LLMs and how enterprises can benefit from them.

Read Ebook

Part 2: How LLMs Are Unlocking New Opportunities for Enterprises

Explore how traditional natural language processing tasks are performed by LLMs, including content generation, summarization, translation, classification, and chatbot support.

Read Ebook
Architecture diagram of NVIDIA AI Blueprint for telco network configuration.
公司信息
新闻与活动
热门链接