130 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 2026-10-05
stories 22

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

TRL widens template matching for LFM2, LFM2.5, and Gemma 4

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

TRL's chat-template matcher went from exact-match against one stored copy to recognizing the revisions these models actually ship, which unblocks assistant-only loss and tool-calling training on pinned checkpoints.

Recognize the current LFM2 chat template for assistant-only loss huggingface/trl

by albertvillanova

TRL now recognizes the chat template currently shipped by LiquidAI/LFM2-1.2B and the other LFM2 checkpoints that share it, so SFTTrainer with assistant_only_loss=True works again on those models. The template was updated to render assistant tool_calls, and the exact-match check against TRL's stored copy had been rejecting it.

Recognize the other Gemma 4 chat template revisions huggingface/trl

by albertvillanova

TRL recognizes every revision of the Gemma 4 chat template shipped by google/gemma-4-E2B-it, so add_response_schema and GRPOTrainer with tools work on a pinned revision rather than only the single revision TRL stored. The repo has shipped four other revisions.

Recognize the current LFM2.5 chat template huggingface/trl

by albertvillanova

The current LiquidAI/LFM2.5-230M chat template is now accepted as well, restoring add_response_schema and GRPOTrainer-with-tools on that model after the template changed post-storage. All three land under the same effort to stop identifying models by exact match alone.

Count dashboard page views huggingface/transformers-ci

by tarekziade

Grafana OSS counts only route patterns like /d/:uid/:slug, so there was no way to tell which dashboards get visited. Each dashboard's hidden loader panel now sends GET /page-view?d=<uid> once per visit, and the trace-exporter returns 204 while incrementing pytest_trace_exporter_page_views_total{dashboard}; the label is uid-validated and capped at 50 series with an "other" bucket.

chore(deps): bump huggingface/doc-builder/.github/workflows/build_main_documentation.yml from ef23fb82532fe347298732dc81dacad2f8eff961 to 9d137cbeab5c75e8093b5c0036f83a429cf878fb huggingface/OpenEnv

by dependabot

The rest of the day is dependency and CI housekeeping: OpenEnv bumped the doc-builder build_main_documentation workflow pin, and tornado moved from 6.5.8 to 6.5.9 in /envs/coding_env.

Quick answers

What shipped in Hugging Face on October 5, 2026?
TRL's chat-template matcher went from exact-match against one stored copy to recognizing the revisions these models actually ship, which unblocks assistant-only loss and tool-calling training on pinned checkpoints. In total, 6 commits and 16 pull requests landed.
Who contributed to Hugging Face on October 5, 2026?
3 developers shipped this update, including dependabot, tarekziade, and albertvillanova.
What were the notable Hugging Face updates?
Recognize the current LFM2 chat template for assistant-only loss, Recognize the other Gemma 4 chat template revisions, and Recognize the current LFM2.5 chat template.