Feature Request Hugging Face Docs

[ueoo]

The Hugging Face 🤗 libraries, such as Hub, Transformers, Diffusers, Accelerate, Datasets, and Gradio, are quite useful and popular 🔥 among the AI community.
Could anyone try to incorporate these docs into Dash?
Thanks!

[sxysxy]

I've tried to build docset on latest formal version of transformers(v4.57.3). Huge changes have been made from current latest docset (v4.5.0), It's hard to build the docset on this version of transformers, I have to write some scripts to fix the directory structure, and hyperlinks in the html files, make the dsidx indexing works.

Still, I'm facing these issues:

• Missing all CSS styles.

• A few hyperlinks are broken.

• Searching works

• Missing CSS styles

[sxysxy]

Hi Dash maintainers,

I’ve been experimenting with updating the Transformers docset to a much more recent version of the library. Over the past few years, Transformers has added a very large number of new models, classes, and APIs, while the current transformers Dash docset is still based on transformers v4.5.0, which is now quite outdated and misses a lot of commonly used content.

Due to changes in how Hugging Face builds their documentation (it is now a JavaScript-driven SPA), it is no longer straightforward to generate a “traditional” static docset. However, I’ve managed to build a best-effort offline docset for newer Transformers versions with the following properties:
- Search/indexing works correctly in Dash (models, classes, methods are searchable).
- Most internal documentation links work (there are still a small number of broken links).
- The content is readable, though not beautiful due to missing CSS styles.

I can provide:
- The patch scripts used to post-process the generated documentation.
- A reproducible method for building a docset from recent Transformers documentation.

Before preparing a full PR, I would like to ask whether the Dash community would be open to accepting such a docset, clearly documented as a best-effort / experimental update, prioritizing up-to-date content and searchability over visual fidelity.

From a practical point of view, this newer docset is significantly more useful than the current one, despite the styling limitations.

Thanks for your time, and I’m happy to adjust the scope or presentation based on your feedback. @Kapeli

[Kapeli]

I think that would be fine. Have you tried using the Docset Generator from Dash's Settings > Downloads > Docset Generator? It's able to grab JS-heavy pages quite well.

[sxysxy]

I think that would be fine. Have you tried using the Docset Generator from Dash's Settings > Downloads > Docset Generator? It's able to grab JS-heavy pages quite well.

Hi, Thanks for the reminding of surprising docset generator. I just tried the docset generator for transformers v4.57.3, using local static html files. The main issues are:

• 1. Only class entities could be automatically indexed via 'Basic index', the method entities are lost. Maybe 'Custom index' could fix. I used to adopt 'doc2dash' tool and my patching script, they could build method level index.
• 2. It seems that all internal links are all broken, for example, clicking 'Qwen3MoE' from 'Qwen3' page will get 'Page not found'
• 3. Also missing CSS styles.
• 4. I also tried 'Download website' method and reviewed the 'Generation Log', I found the docset could only find the index page(https://huggingface.co/docs/transformers/v4.57.3/en/index) and could not dive into more pages. Maybe there's some anti-spider mechanism.

I just started trying out this docset generator, so maybe I’m not using it well enough yet. For now from my side, it seems that the docset I generated using my previous method is slightly better.

[Kapeli]

I'm able to get the Docset Generator to download the docs. Make sure you set "Download pages matching:" to huggingface.co/docs/transformers/v4.57.3/en/*.

However, it fails later on because of reaching a rate limit. I'll modify the docset generator to follow RateLimit-Policy headers and will get back to you when it's ready.

[Kapeli]

DocsetGenerator is now able to handle rate limiting. However, the HuggingFace website has quite a restrictive rate limit, so please note that cloning the docs will take some time.