This week was really exciting because we had two project presentations lined up — one from our classmates on Preswald, and one from our team on Huggingface.

Preswald Presentation

Our classmates introduced us to Preswald, a tool focused on building data visualization dashboards using Python and HTML. It’s structured around an Open Core model, where the base software is free, but customizations are paid.

They talked about the community structure around Preswald — including a small but active subreddit and responsive maintainers. However, they also pointed out challenges like a poorly managed Slack channel and sluggish issue/PR responses.

In terms of their actual contributions, the highlights were:

• Implementing LaTeX support
• Working on sidebar functionality (which was tricky due to token handling)
• Building an interactive chat interface based on their documentation
• Adding a debug panel for better development and testing

They also reflected honestly on the difficulties they faced: lack of documentation, needing to learn the underlying tech from scratch, fast changes to the base code, and some frustration over the limited community support.

A cool tip they shared was how important it was to get on calls with the developers directly, and that a contributing guide would have made onboarding a lot smoother.

---

Huggingface Presentation

Our team presented our work on Huggingface 🤗, one of the biggest names in the open-source AI space.

We walked through:

• What Huggingface does (open-source models like Transformers and Diffusers, huge datasets like Wikipedia and COCO)
• Our contribution statistics (22 contributions: 11 PRs and 11 issues across features, bug fixes, documentation, community support, etc.)
• How we communicated on Slack, coordinated our pull requests, and divided up work

Each of us shared our individual contributions:

• Haocheng worked on bug fixes (like correcting position ID increments), documentation fixes, and feature improvements for Qwen2 MoE models.
• Yufeng focused on bug fixes, type consistency improvements, and issues related to KV cache generation.
• Haochen also contributed bug fixes and clarified tricky logic in the code.
• Minjun handled documentation, added tests for new cache mechanisms, and introduced resume checkpoint functionality for ClearML.

For our group contribution, we highlighted how we implemented a time-based evaluation, saving, and logging strategy for the Huggingface Trainer — making it more consistent across different hardware setups.
We even created three different test scenarios to verify that the time-based strategies worked reliably!

---

Final Thoughts

It was really interesting to see the contrast between the two projects:

• Preswald is smaller, scrappier, and still building its foundation, while
• Huggingface is a massive, polished open-source community with structured contribution processes.

Both presentations made it clear that no matter the size of the project, successful contributions rely on:

• Understanding the codebase
• Clear communication
• Careful testing
• And perseverance in navigating project challenges

Really proud of what everyone achieved this week! 🎉