Keynote: To Nemotron and Beyond: Building the Open Source AI Ecosystem
summary
NVIDIA claims open weights are driving an agentic open-source stack and lists concrete outputs: 650 open models and 250 open datasets on Hugging Face, LangChain at ~80M downloads, and a new NeMo family (Nemotron / NeMo 3 lightning) with a 30B/3B-active configuration that runs on a single H100 and competes with much larger models after fine-tuning. The openness score highlights missing methodology and pretraining provenance, and demonstrates that weights plus recipes enable task-specific post-training and faster customization.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:05Open source ecosystem
- 2:08Open models democratization
- 3:55NVIDIA’s open source AI
- 4:59Open models accelerate innovation
- 6:27Rising cost of reasoning
- 7:10True open source vs open weights
- 9:01AI Openness Index gaps
- 10:05Beyond model size
- 11:32Building open AI ecosystem
related talks
16mKeynote: Open Source AI Everywhere
13mKeynote: Introductory Remarks
11mKeynote: Open Source Stack Empowers the AI Era
12mKeynote: Build Accelerated LLM Applications with Open Source Tools from NVIDIA
11mKeynote: Building an Interoperable Ecosystem for Generative AI
14mKeynote: Building AI with Open Source & Hugging Face
27mKeynote: Open Source in the Age of AI- 14mKeynote: Building AI with Open Source & Hugging Face