Weekly Hugging Face trending digest

Every Monday at 9am, scan the Hugging Face Hub (read-only) for what is trending and newly published in the topics I follow, and email me a skimmable digest. Follow these steps exactly:

1. Use these topics unless I have edited them below: a model task of "text-generation" and a dataset topic of "embeddings" (dataset tag "task_categories:sentence-similarity"). Treat each topic independently.
2. For models, run model_search with the `task` parameter set to the pipeline task (e.g. "text-generation") and optionally the `library` parameter (e.g. "transformers"), sorted by trendingScore (descending). Run it a second time sorted by createdAt (descending) to catch freshly published repos. Keep the top ~5 from each pass and de-duplicate by id. Note: model search filters only by task and library, not by arbitrary tags.
3. For datasets, run dataset_search filtered by the relevant tags (e.g. "task_categories:sentence-similarity"), once sorted by trendingScore and once by lastModified (descending). Keep the top ~5 from each, de-duplicated.
4. For Spaces, run space_search with a plain-text query describing the topic (e.g. "text generation demo"). This is a semantic relevance search only — it cannot sort by recency, downloads, or likes, and cannot filter by tag or task. Rank the returned hits yourself by likes and trendingScore and keep the top ~3. Frame these as "relevant Spaces", not "newest".
5. For each item, report: the name/id, what it is (task/library or a short description), why it is notable (downloads, likes, trendingScore, and how recently it was published via createdAt or updated via lastModified), and a direct link (https://hf.co/<id>, https://hf.co/datasets/<id>, or https://hf.co/spaces/<id>).
6. "Trending" means the Hub's own trendingScore (a current snapshot weighting recent activity), and "new" is inferred from createdAt — you are reading a point-in-time view, not a true week-over-week diff. Do not claim an item is "new since last Monday".
7. Stay strictly read-only: only search and read repo details. Never publish, like, or modify anything. Do not invent ids, counts, or links — use only what the tools return.
8. Email me the digest grouped as Models / Datasets / Spaces, each item one or two lines, most notable first. If a search returns nothing for a topic, say so plainly for that section instead of guessing or padding with unrelated results.

Hugging Face digest — week of 29 Jun 2026

Two strong new text-generation models trending this week; embeddings datasets quiet.

Models (text-generation)
- aurora-ai/Aurora-12B-Instruct — transformers, instruction-tuned 12B. Trending #4; 38k downloads, 1.2k likes; published 24 Jun. https://hf.co/aurora-ai/Aurora-12B-Instruct
- nordlys/Lumen-Mini-3B — compact 3B for on-device chat. 9.6k downloads, 410 likes; published 27 Jun. https://hf.co/nordlys/Lumen-Mini-3B

Datasets (embeddings / sentence-similarity)
- vectorlab/web-pairs-v2 — 2.1M sentence pairs for retrieval training. 540 likes; updated 21 Jun. https://hf.co/datasets/vectorlab/web-pairs-v2

Spaces (relevant, ranked by likes/trending)
- playground/text-gen-arena — side-by-side model comparison demo. 2.3k likes, trending. https://hf.co/spaces/playground/text-gen-arena

Note: Spaces are a relevance match for your topic, not a "newest" feed.

More Research templates