Salesforce与Nvidia联手推出首个推理模型Koa，基于Nvidia开源的Nemotron模型微调而成1。该模型专为销售、营销和客户支持任务优化1，将通过Salesforce Agentforce平台的AI"网关"自动路由1。
Koa模型在成本和安全性方面具有优势。Salesforce采用合成数据而非真实客户数据进行训练1，相比Claude或ChatGPT具有更低的token成本和更高的token效率1。Salesforce AI执行副总裁Jayesh Govindarajan表示，"推理一直是我们依赖前沿模型提供商的事情。直到现在"1。Nvidia生成式AI软件企业副总裁Kari Ann Briski补充道，"这是你需要的三位一体的东西：主权AI、首个token的时间、高效推理"1。
此外，Salesforce同时宣布与Anthropic的ClaudeForce合作1。
Salesforce has unveiled Koa, its first reasoning model developed in partnership with Nvidia, marking a significant move to reduce reliance on proprietary AI systems from frontier labs.1 The model is built on Nvidia's open-source Nemotron foundation and has been fine-tuned specifically for sales, marketing, and customer support tasks.1 Koa represents a direct alternative to commercial offerings, delivering lower token costs and improved efficiency compared to models like Claude or ChatGPT.1
Rather than relying on external model providers for reasoning capabilities, the company trained Koa using synthetic data instead of customer information, enhancing data security and privacy protections.1 The model will be automatically routed through an AI "gateway" within Salesforce's Agentforce platform, streamlining deployment for users.1 Jayesh Govindarajan, Salesforce's executive vice president of AI, stated: "Reasoning has always been something we've been dependent on frontier model providers for. Until now."1
Kari Ann Briski, vice president of generative AI software at Nvidia, described the initiative as "the trinity of things you need: sovereign AI, time to first token, efficient reasoning."1 Notably, Salesforce simultaneously announced a separate partnership with Anthropic to offer ClaudeForce, indicating a multi-model strategy for enterprise customers.1
评论
还没有评论，欢迎留下第一条。