README.md

metadata

title: README
emoji: 🏃
colorFrom: red
colorTo: blue
sdk: static
pinned: false

We are a community of researchers from different countries and institutions interested in studying the emerging open source AI ecosystem. Learn more about our areas of interest below and join us to uncover the activity and impact of different initiatives focused on openness in AI.

[Complete] Quantitative Analysis of Hugging Face Hub

• Using activity data from Hugging Face Hub datasets, models and spaces, we ran a three-part quantitative analysis of development activity, finding that:
  • activity is extremely imbalanced between repositories; with over 70% of models having 0 downloads, while 1% accounting for 99% of downloads
  • the community has a core-periphery structure, with a core of prolific developers while the large majority (89%) of developers are "islands", primarily working alone.
• Link to Paper: The AI Community Building the Future? A Quantitative Analysis of Development Activity on Hugging Face Hub

[Ongoing] Model Tree Lineages from Base Models to Derived Models

• We are looking into what kinds of downstream activity are most prevalent for different kinds of base foundation models.
• What kinds of derivations (e.g. fine tuning, quantizing) are most popular?

[Ongoing] Cartography of Collaboration in the Open Source AI Ecosystem

• We are investigating how open source AI artefacts such as open datasets and open-weight models are built and who contributes to this process across the model building pipeline, including during the pre-release and post-release phases.
• Our research aims to map collaboration practices from pre-release to post-release across diverse organisational contexts, including individuals, grassroots initiatives, research institutes, startups, and large technology companies, and geographies.