{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreih7kay6stexnjdfutggvah7pcdq2ckimd4vjjcjug62b5dqgvn4la",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mix4jp4tpvf2"
  },
  "path": "/t/where-do-i-host-an-optimizer/175063#post_2",
  "publishedAt": "2026-04-07T23:38:47.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "the TRL GitHub repository",
    "Models, Datasets, and Spaces",
    "Space",
    "Show and Tell",
    "Dataset repo",
    "Collection",
    "PRO",
    "Blog Explorers",
    "new-blog"
  ],
  "textContent": "As for the Optimizer, I don’t think there’s any problem with putting it in the HF repository, but that just lets you store Python files and such, so it’s not very practical…\nAlso, if you want to integrate your optimizer into the Hugging Face ecosystem directly, the best approach is probably to commit it to the TRL GitHub repository.\n\nAlternatively, the following method is commonly used:\n\n* * *\n\nHF doesn’t really have a dedicated “optimizer” section. The Hub’s first-class repo types are Models, Datasets, and Spaces, so for an optimizer library I’d usually keep the **code/package on GitHub + PyPI** and use HF for the parts that fit naturally.\n\nConcretely, I’d suggest:\n\n  * **Space** for an interactive demo or benchmark explorer\n  * **Show and Tell** for the announcement / discussion thread\n  * **Dataset repo** only if you want to publish benchmark traces or reproducibility data\n  * **Collection** to group the Space + dataset + any paper in one place\n  * **Community Blog** as a later long-form writeup, not the primary home\n\n\n\nSo if it were me: **GitHub/PyPI as the canonical home, HF Space as the demo, then a Show and Tell post linking everything**.\n\nCommunity Blog could also work if you want a longer technical article, but I’d treat that as a follow-up: HF says **PRO** includes Community Blogs, and the current flow goes through **Blog Explorers** / **new-blog**.",
  "title": "Where do I host an optimizer?"
}