Together AI 发布 FlashAttention-4 等七项创新，强化 AI 云全栈能力
原标题：Key research and product announcements at the AI Native Conf
AI 摘要
Together AI 在首届 AI Native Conf 上发布了七项研究和产品，涵盖内核、强化学习和算法推理优化。其中 FlashAttention-4 针对 NVIDIA Blackwell GPU 优化，速度比 Triton 快 2.7 倍；Together Megakernel 将实时语音代理的首 token 延迟从 281ms 降至 77ms；together.compile 自动化内核优化，使图像生成提速 41%。此外，Together 还推出了强化学习 API 和 ThunderAgent，以加速 RL 训练和代理工作负载。
正文节选
Together Research is announcing FlashAttention-4, Reinforcement Learning API, ThunderAgent, ATLAS-2, and more at AI Native Conf. The AI Native Cloud is more than a positioning statement. It is a full-stack AI cloud that is purpose-built for AI-natives by researchers and engineers who have delivered foundational AI work such as FlashAttention and ThunderKittens. The same people who published that research are the ones running the production systems our customers, such as Cursor and Decagon, depen