Remove preload_from_hub documentation from spaces config reference

[Pierrci]

Deprecated in favor of mounting volumes

related slack thread

---

Note

Low Risk
Low risk documentation-only change that removes guidance for a deprecated config option; no runtime behavior is affected.

Overview
Removes the preload_from_hub section from the Spaces configuration reference (spaces-config-reference.md), including its YAML examples and caveats, reflecting that this option is no longer documented/encouraged.

^{Reviewed by Cursor Bugbot for commit 476d816. Bugbot is set up for automated code reviews on this repo. Configure here.}

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Wauplin]

As someone wisely told the team lately:

please remember to ALWAYS link to any slack/email/whatever available context in your PRs, it's exhausting having to figure out where this or that was discussed - thank you for your attention!

So I'm cross-posting this internal slack thread https://huggingface.slack.com/archives/C0256NXF0A2/p1776713912103899

😇

[pcuenca]

I can't see the Slack thread that was mentioned, but cc @merveenoyan because she was proposing this as a solution yesterday.

Is there a similar configuration for buckets to be mounted? (i.e., something that represents what this UI exposes)

[Wauplin]

I can't see the Slack thread that was mentioned

Yes sorry it's a private convo (nothing much in it though)

Is there a similar configuration for buckets to be mounted?

Not as yaml metadata no. It's handled in the Space settings like repo visibility or Space hardware

[julien-c]

we could maybe keep the metadata and support it through mounted repos transparently for the user?

[Pierrci]

why not, but maybe scrap the portion that targets specific files, cc @XciD

  - warp-ai/wuerstchen-prior text_encoder/model.safetensors,prior/diffusion_pytorch_model.safetensors
  - coqui/XTTS-v1
  - openai-community/gpt2 config.json 11c5a3d5811f50298f278a704980280950aedb10

[julien-c]

yes indeed^

[Pierrci]

i'm a bit worried about having two different sources of truth though (readme metadata and spaces settings), and the conflicts that come with it (and more importantly, users trying to debug them)

[Wauplin]

maybe let's wait to figure out exactly what we want to promote instead of preload_from_hub before removing it from the docs? (especially since hf-mount needs some fixes for mmap loading apparently + deserves some more benchmarks to check hf-mount is worth it in this use case)