本文来源:https://theagenttimes.com/articles/ian-buck-outlines-on-prem-experiment-planning-agents-at-sc26-155297cd
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
据Brian Caulfield 11月17日在NVIDIA博客上发布的文章称,NVIDIA超大规模和高性能计算副总裁Ian Buck周一在芝加哥举行的SC26上拉开了该公司本周活动的序幕,介绍了能够在研究人员自己的系统上规划实验、运行代码和检查结果的软件代理。据报道,巴克表示,在他提到的阿贡国家实验室使用新工具包构建的代理,仅需一个提示,即可规划和测试假设,选择要运行的模拟,并确定两个新的结合位点。
11月17日的报道称,巴克在演讲伊始回顾了他20年前在佛罗里达州坦帕举行的SC06教程上首次公开演讲CUDA的情况。他说,英伟达目前提供超过1000个CUDA-X库。根据11月17日的报道,CUDA是NVIDIA于2006年11月宣布并在SC06上向科学界推出的平台,目前在整个计算栈上有超过1100万名开发者在其基础上进行开发。16 NVIDIA 文章,作者:Brett Murray。
根据演讲摘要,巴克表示:“一如既往,计算和新科学技术栈的前沿将是模拟,但将与人工智能和未来的量子技术相结合。” 他说,模拟仍然是科学计算的基础,“它将永远处于顶端”,而人工智能、量子计算以及现在的代理将与之协同工作。Buck表示:“[代理]彻底改变了我们与超级计算机、设备、仪器的交互方式,以及我们开展研究的方式。”
据该博文称,在这次演讲中,Buck还宣布了用于高性能计算的Nemotron for Science、Rubin HPC GPU和扩展的Vera CPU平台,并详细介绍了与IBM和费米国家加速器实验室(Fermi National Accelerator Laboratory)开展的一项新的量子计算合作。
面向研究人员计算机的科学版Nemotron
根据NVIDIA的博文,Buck解释说,他为科学代理描述的软件包Nemotron for Science将Nemotron 3.5 Super VL开放模型与一个新的开放化学数据集以及针对生命科学和化学的后训练方案相结合。该文章指出,研究人员可以在DGX Spark、工作站或自己的超级计算机上本地运行该模型。“您拥有自己的数据。您拥有该模型。”Buck 表示。
据该报道称,巴克表示,使用该工具包构建的阿贡实验室智能体仅需一个提示,即可规划和测试假设,选择要运行的模拟,并识别出两个新的结合位点。根据同一篇报道,他随后重点介绍了其他已投入使用的智能体。
该帖子指出,来自瑞典皇家理工学院(KTH Royal Institute of Technology,本文中提及的机构)的代理能够自行修复损坏的GROMACS设置,并且在一台DGX Spark上运行一整夜,就修复了81%的测试用例。在文章中提到的SETI研究所的例子中,艾伦望远镜阵列的一个代理使用一行提示绘制了一条甲醇谱线,巴克解释说,该代理随后自行重新调整了多普勒频移,英伟达写道。据该文章称,NVIDIA的Nsight AI Porting Agent(该文章称之为NVIDIA的新代理)将一个例程从天气研究和预报模型移植到GPU上,据巴克介绍,该例程使用约1.5亿个令牌,速度提高了5倍以上。
11月17日的帖子显示,该回顾介绍了两个结合位点、81%的数字、甲醇折线图以及天气研究和预报加速,这些内容均是Buck演讲的一部分。
据11月17日的文章所述,巴克解释说,代理依赖CPU在模型步骤之间运行工具调用。“这正是代理所需要的,”他说。“它们需要一个能够运行、评估、运行工具、执行操作、收集结果的CPU。”
根据11月17日的帖子,对于自主研究系统的操作人员,NVIDIA的账户介绍了可以在研究人员自己的系统上规划实验、运行代码和检查结果的软件,并表示研究人员可以在DGX Spark、工作站或他们自己的超级计算机上本地运行Nemotron for Science。该帖子记录了一个GROMACS代理,它在一台DGX Spark上进行一次过夜运行,修复了81%的测试用例;一个Allen望远镜阵列代理,它根据一行提示绘制了一条甲醇线;以及NVIDIA的Nsight AI移植代理,Buck表示,该代理将天气研究和预报例程的速度提高了5倍以上。Buck表示:“您拥有自己的数据。您拥有该模型。”他解释说,代理需要一个能够运行该工具、执行操作并在模型步骤之间收集结果的 CPU。两个结合位点、81% 的过夜数据和超过 5 倍的提速是该地址的帖子记录的结果。
[来源] blogs.blog ×3 · 已记录引用 blogs.nvidia.com →
[来源] nvidianews.news ×3 · 记录的引用 nvidianews.nvidia.com →
Hugging Face 将推理提供商路由切换至 OpenRouter 代理平台
NVIDIA详细介绍用于车载推理代理平台的Alpamayo 2 Edge
Claude Playbook建议开发者从Opus 5.5代理平台开始
Amazon Bedrock 为 Agent Calls Agent 平台新增 Kimi K3.1 Pro 高速版
红杉转发了 Ricursive 关于 650 亿美元美国公开权重代理编码领先代理平台的说法
由《The Agent Times》平台、研究与广告技术编辑Roman Bietin发布。所有事实陈述均根据我们的编辑标准注明出处。
《代理人时报》是一家独立的新闻出版物,报道人工智能领域的新闻。
代理经济——平台、开源、研究、商业
每天早上在您的收件箱中收到类似的故事
关于人工智能代理经济的每日电子邮件——内容精辟、易于阅读,每天早上发送。专为希望了解不止一个平台头条新闻的读者打造。
本文来源:https://theagenttimes.com/articles/ian-buck-outlines-on-prem-experiment-planning-agents-at-sc26-155297cd
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, opened the company’s week at SC26 in Chicago on Monday with software agents that can plan experiments, run code and check results on researchers’ own systems, according to a Nov. 17 NVIDIA blog post by Brian Caulfield. Buck said an agent built with the new toolkit at Argonne National Laboratory, the lab he cited, took a single prompt, planned and tested hypotheses, chose which simulations to run and identified two new binding sites, the post reports.
Buck opened the talk by looking back to his first public talk on CUDA at an SC06 tutorial in Tampa, Florida, 20 years ago, the Nov. 17 account states. He said NVIDIA now offers more than 1,000 CUDA-X libraries. CUDA, the platform NVIDIA announced in November 2006 and brought to the scientific community at SC06, now has more than 11 million developers building on it across the computing stack, according to a Nov. 16 NVIDIA post by Brett Murray.
“The frontier of computing and the new scientific stack is going to be simulation, as always, but combined with AI and the future of quantum,” Buck said, according to the address recap . Simulation remains the foundation of scientific computing, “which will always be on the top,” he said, with AI, quantum and now agents working alongside it. Buck said, “[Agents] have been just a game changer in how we interface with our supercomputers, with our devices, our instruments and how we can do the research.”
In that address, Buck also announced Nemotron for Science, the Rubin HPC GPU and an expanded Vera CPU platform for high-performance computing, and he detailed a new quantum-computing collaboration with IBM and Fermi National Accelerator Laboratory, the post says.
Nemotron for Science on the Researcher’s Machine
Nemotron for Science, the package Buck described for science agents, combines the Nemotron 3.5 Super VL open model with a new open chemistry dataset and post-training recipes for life sciences and chemistry, he explained, according to the NVIDIA post . Researchers can run it locally on a DGX Spark, a workstation or their own supercomputer, the post says. “You own your data. You own the model,” Buck said.
Buck said the Argonne agent, built with the toolkit, took a single prompt, planned and tested hypotheses, chose which simulations to run and identified two new binding sites, the post reports. He then highlighted other agents already in use, according to the same account .
The post says an agent from KTH Royal Institute of Technology, the institution named for this case, repairs broken GROMACS setups on its own and, in one overnight run on a single DGX Spark, fixed 81% of test cases. An agent at the Allen Telescope Array, in the post’s SETI Institute example, took a one-line prompt to plot a methanol line, and Buck explained that it then retuned for Doppler shift on its own, NVIDIA wrote . NVIDIA’s Nsight AI Porting Agent, which the post calls NVIDIA’s new agent, ported a routine from the Weather Research and Forecasting model to GPUs, Buck said, speeding it up more than 5x using about 150 million tokens, according to the post .
The recap presents the two binding sites, the 81% figure, the methanol-line plot and the Weather Research and Forecasting speedup as part of Buck’s address, the Nov. 17 post shows.
Agents lean on CPUs to run tool calls between model steps, Buck explained, the Nov. 17 post says. “That’s exactly what agents need,” he said. “They need a CPU that can run, evaluate, run the tool, perform the operations, gather the results.”
For operators of autonomous research systems, NVIDIA’s account describes software that can plan experiments, run code and check results on researchers’ own systems, and says researchers can run Nemotron for Science locally on a DGX Spark, a workstation or their own supercomputer, according to the Nov. 17 post . The post records a GROMACS agent that fixed 81% of test cases in one overnight run on a single DGX Spark, an Allen Telescope Array agent that plotted a methanol line from a one-line prompt, and NVIDIA’s Nsight AI Porting Agent, which Buck said sped a Weather Research and Forecasting routine more than 5x, the account states. Buck said, “You own your data. You own the model,” and explained that agents need a CPU that can run the tool, perform the operations and gather the results between model steps. The two binding sites, the 81% overnight figure and the more-than-5x speedup are results the post records from that address.
[SOURCE] blogs.blog ×3 · recorded citation blogs.nvidia.com →
[SOURCE] nvidianews.news ×3 · recorded citation nvidianews.nvidia.com →
Hugging Face Switches Inference Providers Routing to OpenRouter Agent Platforms
NVIDIA Details Alpamayo 2 Edge for Onboard Vehicle Reasoning Agent Platforms
Claude Playbook Tells Builders to Start on Opus 5.5 Agent Platforms
Amazon Bedrock Adds Kimi K3.1 Pro Highspeed for Agent Calls Agent Platforms
Sequoia Reposts Ricursive Claim of 65B U.S. Open-Weight Agentic-Coding Lead Agent Platforms
Published by Roman Bietin , Platforms, Research & Ad Tech Editor for The Agent Times. All factual claims are sourced per our editorial standards .
The Agent Times is an independent news publication covering the AI
agent economy — platforms, open source, research, commerce
Get stories like this in your inbox every morning
A daily email on the AI agent economy — sharp, readable, and sent each morning. Built for readers who want more than a single platforms headline.
