This week, we discussed the paper by Eric S Raymond The Cathedral and the Bazaar and watched the talk by Claude Warren The Cathedral, the Bazaar, and the Coffeehosue. Also, I did some research on the Open Source Program Offices (OSPO) at Meta and NVIDIA. Click below to see my thoughts!

Bazaar, Coffeehouse and Huggingface

For those who are confused, the Cathedral, the Bazaar, and the Coffeehouse refer to 3 different models of open source software development:

• The Cathedral represents a centralized development structure, where a (almost) fixed group of developers build the software in a controlled environment.
• The Bazaar refers to decentralized and collaborative environment, and a diverse community of developers build the software together.
• The Coffeehouse is somewhat similar to the concept of Bazaar, which both focus on a decentralized and collaborative environment. However, Bazaar is more focused on the behavior of development itself (push & pull, buy & sale), whereas Coffeehouse is more about discussion and exchanging thoughts.

Some ideas that I really relate to in Eric’s paper are:

“Good programmers know what to write. Great ones know what to rewrite.”

In my own words, good programmers are able to implement some features from scratch, yet great programmers can understand a large, complex, and collaborative codebase and rewrite a small portion to achieve the features desired. This saves the pain of building wheels over and over again, and is an essential skill in any open-source (or large-scale) software development.

“Treating your users as co-developers is your least-hassle route to rapid code improvement and effective debugging.”

I believe “treating your users as co-developers” means you need to guarantee the readability, consistency, and extensibility in your codebase. With these qualities guaranteed, the users can trace the source of bugs by themselves and directly propose solutions. On the contrary, if the codebase turns out to be very obscure, the users may only be able to report the problems they encountered but not able to attribute the problems to certain portions of the code, which takes longer for the developers to debug and improve the code.

I believe the Bazaar and the Coffeehouse models are very relevant to the Huggingface open source projects. The Huggingface community is like a Bazaar and a Coffeehouse because the code is very structured, easy for contributor from all over the world to study and contribute to. Also, Huggingface has a very active forums, with well-categorized topic sections.

The Coffeehouse

Personally, I’m a huge fan of coffee, and I like the idea of Coffeehouse in Open Source as well. I believe it’s an appropriate augmentation to the Bazaar, because developers from all over the worlds should not only exchange code, but also it’s necessary to actively communicate with each other to ensure they share the same values and the same ideas of the style of the codebase to develop.

The role of OSPO

I did some research on the OSPOs of Meta and NVIDIA:

• Some people working at Meta Open Source are Jon Janzen, Dmitry Vinnik, Cami Williams, and AJ Bush. One open source project by Meta that I’m most familiar with is the Llama family of language models, which is one of the canonical base model for open source development of NLP projects and academic research at this moment. I believe the Llama project as well as its development team greatly enhanced Meta’s influence in the Natural Language Processing and Computer Vision communities.
• According to Github, it seems some active open source developers at NVIDIA are Shanqing Cai, Oleksii Kuchaiev, and Somshubra Majumdar. Some NVIDIA open source projects I’m familiar with are NeMo, Cosmos, and TensorRT Model Optimizer, which are all generative AI projects. NVIDIA also provides a myriad of CUDA related projects, which help more people learn about CUDA and use CUDA in their own development. I believe OSPO is vital for NVIDIA to keep their dominance in the GPU market, as they provided abundant NVIDIA-GPU-friendly software support through OSPO, which makes more people choose NVIDIA GPUs.