The Model Openness Framework: Promoting Completeness and Openness for Reproducibil...
summary
A framework that classifies models by completeness and binary openness and lists 16 artifacts (architecture, weights, eval data, model card, etc.) produces a machine-readable moff.json index file and a web tool that issues badges and a scoreboard. Three classes (class one = full open science, class two = reproducible, class three open model = architecture plus weights) guide licensing choices; plan is to test every “supposedly open” model, start with Alther AI and R wkv, and launch the tool at AidDev.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:50Openness in AI challenges
- 4:59Open science in AI
- 7:07Open data in models
- 9:35Completeness vs openness
- 12:26Licensing in AI models
- 14:33Model Openness Framework
- 17:42Model openness tiers
- 21:11Model Openness Tool
- 22:17Benefits for model producers and consumers
- 25:57Framework limitations
- 27:44Call to action
- 29:48Q&A
related talks
26mIntroducing the Model Openness Framework: Achieving...
29mFostering Responsible AI:Empowering Openness and Community Collaboration
40mThe Generative AI Commons: Unlocking Potential Through Collaboration and Transpare...
32mOperationalizing Openness: Standardizing AI Model Supply Chains With the Model Openness Framework
37mIs Your AI Truly Open ? Cutting Through the Hype
22mThe Role of Openness in Responsible AI
39mFrom OpenAI to Opensource AI: Navigating Between Commercial Ownership and Coll...
40mUnleashing Collective Genius: Building GenAI Through Open Collaboration