112 wires and counting

$ follow Hugging Face

Keep up with Hugging Face in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-05-17
stories 14

© 2026 RepoJournal Home Showcase Explore How it works Privacy

$ the-wire · showcase

SWIFT TOKENIZERS FIXED FOR UNICODE, TRANSFORMERS CI PATCHED, MONGOKU SHIPS SCHEMA AUDIT

By RepoJournal · Filed · About Hugging Face · Composed from the cited sources · methodology

Swift-transformers shipped three critical tokenizer fixes that resolve grapheme cluster bugs breaking emoji and combining marks across Unigram, BPE, and BasicTokenizer [ref:8] [ref:9], while transformers CI dodged an E2BIG argument limit blowup on large PRs [ref:1].

The Swift fixes address a fundamental mismatch: SentencePiece vocabularies index by Unicode scalar, but Swift's Character type operates on extended grapheme clusters, causing emoji like '1️⃣' and combining marks in Thai, Devanagari, and Japanese to fail tokenization [4] [5] [6]. These aren't edge cases - they're vocab coverage holes that would silently produce wrong tokens in production. Transformers sidestepped a CI disaster by refetching PR files in-script instead of piping them through environment variables, which was hitting the kernel's MAX_ARG_STRLEN limit on PRs with large patches [7]. Mongoku merged a schema auditing feature with a medium-risk warning: new collection-wide introspection endpoints could surface expensive aggregations that timeout on large datasets [8]. Hub-docs auto-bumped inference provider packages and regenerated docs without incident [9].

Action items

References

  1. [1] Unigram lattice walks Unicode scalars (#352, Bug 3) (#356) ↗ huggingface/swift-transformers
  2. [2] BPE merge by Unicode scalar, not grapheme cluster (#352, Bug 4) (#355) ↗ huggingface/swift-transformers
  3. [3] bugfix(ci): avoid E2BIG in pr_slow_ci_suggestion (#45983) ↗ huggingface/transformers
  4. [4] Unigram lattice walks Unicode scalars (#352, Bug 3) ↗ huggingface/swift-transformers
  5. [5] BPE merge by Unicode scalar, not grapheme cluster (#352, Bug 4) ↗ huggingface/swift-transformers
  6. [6] Strip Japanese voiced-kana marks in BasicTokenizer (#352, Bug 2) ↗ huggingface/swift-transformers
  7. [7] bugfix(ci): avoid E2BIG in pr_slow_ci_suggestion ↗ huggingface/transformers
  8. [8] Add schema auditing endpoints and navigation tab ↗ huggingface/Mongoku
  9. [9] [Bot] Update Inference Providers documentation ↗ huggingface/hub-docs

Quick answers

What shipped in Hugging Face on May 17, 2026?
Swift-transformers shipped three critical tokenizer fixes that resolve grapheme cluster bugs breaking emoji and combining marks across Unigram, BPE, and BasicTokenizer [ref:8] [ref:9], while transformers CI dodged an E2BIG argument limit blowup on large PRs [ref:1]. In total, 6 commits, 7 pull requests, and 1 releases landed.
Who contributed to Hugging Face on May 17, 2026?
4 developers shipped this update, including apocryphx, tarekziade, coyotte508, and HuggingFaceInfra.
What were the notable Hugging Face updates?
Unigram lattice walks Unicode scalars (#352, Bug 3) (#356), BPE merge by Unicode scalar, not grapheme cluster (#352, Bug 4) (#355), and bugfix(ci): avoid E2BIG in pr_slow_ci_suggestion (#45983).