Keynote: Building an Interoperable Ecosystem for Generative AI
summary
Open-source generative AI adoption hinges on concrete interoperability: models must be published to Hugging Face Hub to get noticed, training stacks like GTB nox, accelerate and Megatron-LM target different scales, and one-line conversion scripts that export custom checkpoints to HF formats produce clear adoption inflection points. An evaluation library that integrates local models, cloud APIs and quantization tools implements hundreds of evaluation tasks, runs on AWS, CoreWeave, US government and EU supercomputers, and wins contributors by offering credit and coauthorship.
AI-written summary · about
chapters (10)
- 0:00Interoperable generative AI intro
- 0:44Why interoperability matters
- 1:28Open-source AI ecosystem overview
- 2:12Hugging Face Hub role
- 3:18Training libraries and scale
- 4:21Hardware/software interoperability
- 5:45Evaluation challenges
- 7:08Credit in collaboration
- 8:54Cross-organization collaboration
- 9:16Get involved tomorrow
related talks
44mKeynote: Tracking the Pulse of Generative AI
44mKeynote: Tracking the Pulse of Generative AI
16mKeynote: Open Source AI Everywhere
13mKeynote: State of Open GenAI
44mKeynote: Tracking the Pulse of Generative AI
14mKeynote: Building AI with Open Source & Hugging Face- 13mKeynote: State of Open GenAI
- 14mKeynote: Building AI with Open Source & Hugging Face