NVIDIA高管揭秘如何将模型发布周期缩短一半，还分享了计算资源分配策略
NVIDIA将AI模型发布周期从6-8个月缩短至4-6周。NVIDIA应用深度学习研究副总裁Bryan Catanzaro详细介绍了这一变化。NVIDIA将约30%的Nemotron计算资源用于合成数据生成。视频展示了GPT-3与Muse Glimmer等模型在参数和能力上的对比。
NVIDIA’s has compressed release cycles from every 6–8 months to every 4–6 weeks. Bryan Catanzaro @ct...
NVIDIA’s has compressed release cycles from every 6–8 months to every 4–6 weeks. Bryan Catanzaro @ctnzr , the VP of Applied Deep Learning Research at @NVIDIAAI , breaks down how they are doing it. 00:00 Why AI became core to NVIDIA’s mission 00:55 Infrastructure, data, and algorithms: what drives model progress 01:54 Why NVIDIA invests in the open-source AI ecosystem 03:22 Compute accelerating AI research 04:10 Why ~30% of Nemotron compute goes to synthetic data 04:45 Breakdown of compute used for data and training techniques 05:26 Parameters and capabilities, example: GPT-3 vs. Muse Glimmer 07:03 How NVIDIA allocates compute across research and releases Watch the full video at the link below. Your browser does not support the video tag. 🔗 View on Twitter 💬 2 🔄 0 ❤️ 17 👀 3824 📊 4 ⚡