RepoJournal
Hugging Face @huggingface
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 2026-07-17
stories 60
GitHub RSS Archive Reviews Topics

© 2026 RepoJournal Home Showcase Explore How it works Privacy

$ the-wire · showcase

SENTENCE-TRANSFORMERS FIXES SILENT GRADIENT CORRUPTION ON GPU

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

albertvillanova tomaarsen Andrey Churkin atharvajoshi10 PreethamNoelP Quentin Gallouédec Sayak Paul
8 people shipped this
albertvillanova @albertvillanova 4 cited tomaarsen @tomaarsen 3 cited Andrey Churkin @andrey-churkin 1 cited atharvajoshi10 @atharvajoshi10 1 cited PreethamNoelP @PreethamNoelP 1 cited Quentin Gallouédec @qgallouedec Sayak Paul @sayakpaul yiyixuxu @yiyixuxu

A critical bug in cached losses was silently producing wrong gradients on GPU, and it's fixed now [ref:1].

Sentence-transformers consolidated its fragmented GradCache machinery into a single shared engine and patched a silent gradient corruption bug that affected cross-encoder cached losses on GPU and prompt-training with `Pooling(include_prompt=False)` across all cached losses [1]. This is the kind of bug that ships bad models without warning. In related fixes, the library corrected causal padding validation that was reading the wrong attribute on multimodal processors, causing VLM padding checks to never fire [2], and improved error messages for video metadata passed as sibling keys instead of nested under the video modality [3]. TRL is systematically hardening its experimental trainers with comprehensive eval_dataset initialization tests across GRPO-family [4], Online-DPO-family [5], SFT and distillation [6], and preference trainers [7], mirroring coverage already added to core trainers. Diffusers added Cosmos 3 Edge generator support [9] and began shipping Anima img2img pipeline blocks [10], extending modular generation capabilities. Optimum-Intel patched a missing compression parameter that should have been exposed [8].

Action items

References

  1. [1] [`losses`] Consolidate GradCache into one shared engine, fix silently wrong gradients, add `mini_batch_num_tokens` ↗ huggingface/sentence-transformers
  2. [2] Check causal left padding against the attention mask ↗ huggingface/sentence-transformers
  3. [3] Improve the error message when `video_metadata` is passed as a sibling modality key (#3866) ↗ huggingface/sentence-transformers
  4. [4] Add eval_dataset init tests for experimental GRPO-family trainers ↗ huggingface/trl
  5. [5] Add eval_dataset init tests for experimental Online-DPO-family trainers ↗ huggingface/trl
  6. [6] Add eval_dataset init tests for experimental SFT and distillation trainers ↗ huggingface/trl
  7. [7] Add eval_dataset init tests for experimental preference trainers ↗ huggingface/trl
  8. [8] Add missing compression parameter ↗ huggingface/optimum-intel
  9. [9] Cosmos3 edge support ↗ huggingface/diffusers
  10. [10] [Anima] Add img2img pipeline blocks ↗ huggingface/diffusers

$ ls huggingface/reviews/ # the step-back reads

the week in review OpenEnv 0.5.0 ships as typed clients get fixed the month in review Cache-billing leak fixed, and everything from robots to diffusers gets harder to break
Topics: Python AI / ML Full archive → Weekly reviews →

Quick answers

What shipped in Hugging Face on July 17, 2026?
A critical bug in cached losses was silently producing wrong gradients on GPU, and it's fixed now [ref:1]. In total, 31 commits and 29 pull requests landed.
Who contributed to Hugging Face on July 17, 2026?
8 developers shipped this update, including tomaarsen, Quentin Gallouédec, albertvillanova, Andrey Churkin, Sayak Paul, atharvajoshi10, yiyixuxu, and PreethamNoelP.
What were the notable Hugging Face updates?
[`losses`] Consolidate GradCache into one shared engine, fix silently wrong gradients, add `mini_batch_num_tokens`, Check causal left padding against the attention mask, and Improve the error message when `video_metadata` is passed as a sibling modality key (#3866).
♥

$ log @huggingface --recent

Flash and sage varlen prep work under torch.compile dynamic shapes 6 shipped OpenEnv closes soupsieve CVEs in two more envs, TRL reshapes GRPO log-probs 63 shipped OpenEnv 0.5.0 ships new public surface, capture proxy gains an admin key 48 shipped pwc-cli exposes every read-only command as an MCP tool 47 shipped

$ log --all # elsewhere on the wire

@openclaw Cold-start Gateway diagnostics stop crying wolf, doctor repairs plugin drift Agentic Coding Cline desktop bundles shrink, goose unblocks its own clicks

$ ls showcases/

Rails Django Laravel Node.js Go Vercel Vue.js Kubernetes HashiCorp Arch Linux Linux Anthropic

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.