Who contributed to Hugging Face on August 23, 2026?

4 developers shipped this update, including osolmaz, Haoming Song, HuggingFaceInfra, and pcuenca.

What were the notable Hugging Face updates?

fix(control): settle closed Sandbox file actions, fix(control): clean terminal task sandboxes, and fix(control): settle terminal job reservations.

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

Harbor-hf sinks broken Sandbox cleanup

By RepoJournal · Filed 06:58 UTC on August 23, 2026 · About Hugging Face

4 people shipped this

osolmaz @osolmaz 5 cited

Haoming Song @HaomingSong 1 cited

pcuenca @pcuenca 1 cited

HuggingFaceInfra @HuggingFaceInfra

Harbor-hf shipped five fixes overnight that close the gaps where stuck Sandboxes, terminal jobs, and invalid campaign results could quietly block recovery.

Harbor-hf's controller was leaving Sandbox file actions pending forever when a Sandbox closed mid-write, stalling cleanup and recovery; that path now matches command execution [1]. A finished task could also leave an older capacity hold behind, so the reconciler now closes every admitted Sandbox and waits for a terminal close receipt [2]. Completed jobs held their full cost reservations forever, blocking recovery even when spending stayed under the limit; reservations now release once a job is terminal [3]. The control app could fail at startup rebuilding a large Bucket ledger, so replay now uses a fixed worker pool and retries only transient network failures [4]. And campaigns could seal attempts with no real token usage; the service now rejects those and keeps invalid receipts as evidence [5]. Meanwhile, LeRobot fixed a silent argparse abbreviation that broke all three PEFT CLI tests after draccus 0.11 disabled abbreviations, plus a ZeroDivisionError in RL learner checkpoint saving when save_freq is non-positive [6]. Docs clarify that "Higher Hub rate limits" are not tied to organization IP ranges [7].

Action items

→ Update harbor-hf to include PRs #103-#107 before relying on campaign recovery huggingface/harbor-hf [immediate]
→ Pull in lerobot PR #4477 to fix PEFT CLI tests huggingface/lerobot [plan]
→ Review lerobot PR #4479 to guard checkpoint saving against non-positive save_freq huggingface/lerobot [plan]
→ Note the corrected rate limit docs for enterprise users huggingface/hub-docs [monitor]

References

[1] fix(control): settle closed Sandbox file actions ↗ huggingface/harbor-hf
[2] fix(control): clean terminal task sandboxes ↗ huggingface/harbor-hf
[3] fix(control): settle terminal job reservations ↗ huggingface/harbor-hf
[4] fix(control): bound Bucket replay downloads ↗ huggingface/harbor-hf
[5] fix(control): reject invalid campaign results ↗ huggingface/harbor-hf
[6] fix(tests): fix peft cli test ↗ huggingface/lerobot
[7] Clarify "Higher Hub rate limits" (#2721) huggingface/hub-docs

Quick answers

What shipped in Hugging Face on August 23, 2026?: Harbor-hf shipped five fixes overnight that close the gaps where stuck Sandboxes, terminal jobs, and invalid campaign results could quietly block recovery. In total, 28 commits and 10 pull requests landed.
Who contributed to Hugging Face on August 23, 2026?: 4 developers shipped this update, including osolmaz, Haoming Song, HuggingFaceInfra, and pcuenca.
What were the notable Hugging Face updates?: fix(control): settle closed Sandbox file actions, fix(control): clean terminal task sandboxes, and fix(control): settle terminal job reservations.

LEROBOT CUTS FASTWAM OVERHEAD WITH COMPILED INFERENCE

LeRobot's FastWAM policy gets an opt-in compiled path that slashes Python and CUDA kernel-launch overhead on cached action inference.

python 36 shipped

@huggingface Aug 21

Transformers Brings Biology to v5 Main

Hugging Face's flagship repo hauled two protein-folding models onto the v5 mainline, and a wav2vec2 attention fix is ready to unblock your SDPA workflows.

python 33 shipped

@huggingface Aug 20

TRL DROPS ASYNC DISTILLATION, DIFFUSERS GO TENSOR PARALLEL

HuggingFace's training and inference stacks moved in lockstep overnight.

python 22 shipped

@huggingface Aug 19

TWO NEW TOOLCHAINS AND A PATCH THAT MATTERS

Hugging Face's kernels repo brings Intel's next-gen GPU into the fold, while Transformers quietly fixes a cross-attention bug that could have been eating your encoder context.

python 84 shipped

Elsewhere on the wire

AI Agents Aug 23

Cline hubs now drain without dropping work

Cline's new hub drain and upgrade flow means you can restart a hub mid-session and lose nothing, thanks to durable event replay.

ai-agents 18 shipped

Local LLMs Aug 23

SGLang Fixes Pixtral 500s as vLLM Cuts TTFT 25%

The biggest news overnight: SGLang is patching a production-breaking Pixtral bug, and vLLM engineers found a way to shave up to 25% off time-to-first-token on Mamba models.

llms 134 shipped

@CachyOS Aug 23

AIRE: ANANICY RULES FIX SESSION LAUNCHER SCHEDULING

ly-dm was being pinned to the idle scheduler, killing your login session's responsiveness.

infra 74 shipped

Elixir & Phoenix Aug 23

JOSE VALIM FIXES THE CLASUE BUG

Jose Valim shipped two refactors plus the actual fix for that `none` clause compiler issue, and a UTC fast path in DateTime could save you a TZ DB lookup.

elixir 6 shipped

Want every project, not just this one?

Follow @huggingface