Who contributed to Hugging Face on August 21, 2026?

6 developers shipped this update, including dependabot, jiqing-feng, Yih-Dar, Matt, YangKai0616, and danieldk.

What were the notable Hugging Face updates?

Port ESMC and ESMFold2 to Transformers (#46419), Always tie embeddings for LongT5 and Pop2Piano (#47620), and Retry transient network errors (RemoteDisconnected) in github_utils (#48124).

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

Transformers Brings Biology to v5 Main

By RepoJournal · Filed 07:05 UTC on August 21, 2026 · About Hugging Face

6 people shipped this

dependabot @dependabot 1 cited

jiqing-feng @jiqing-feng 1 cited

Yih-Dar @ydshieh 1 cited

Matt @Rocketknight1 1 cited

YangKai0616 @YangKai0616 1 cited

danieldk @danieldk 1 cited

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.

The bio breakthrough lands this morning: ESMC and ESMFold2 are being ported into Transformers [2], lifting protein-folding straight into the framework's core [1]. The port is the verbatim fork code, with v5 convention adaptation coming as follow-up work [1]. Meanwhile, wav2vec2 conformer and SeamlessM4T now dispatch correctly with attn_implementation="sdpa", fixing crashes during from_pretrained [4]. On the reliability front, a LongT5 and Pop2Piano embedding tie fix prevents dropped structural weights [2], and a Retry transient network errors patch finally corrals the flaky RemoteDisconnected CI tracebacks [3]. The kernels repo lays groundwork with a resolver infrastructure [5] that will recursively solve kernel dependencies in coming PRs. Finally, sentence-transformers bumps setup-uv to 10.0.0, disabling automatic caching for sensitive events [6].

Action items

→ Test ESMC/ESMFold2 on v5 main to catch breaking changes early huggingface/transformers [plan]
→ Upgrade wav2vec2/SDPA path to absorb the dispatch fix huggingface/transformers [immediate]
→ Review the kernel resolver API before building on it huggingface/kernels [monitor]
→ Bump setup-uv to 10.0.0 in CI workflows huggingface/sentence-transformers [plan]

References

[1] Port ESMC and ESMFold2 to Transformers (#46419) huggingface/transformers
[2] Always tie embeddings for LongT5 and Pop2Piano (#47620) huggingface/transformers
[3] Retry transient network errors (RemoteDisconnected) in github_utils (#48124) huggingface/transformers
[4] 🚨[wav2vec2] Support attn_implementation=sdpa dispatch ↗ huggingface/transformers
[5] Add kernel resolver infrastructure ↗ huggingface/kernels
[6] Bump astral-sh/setup-uv from 9.0.0 to 10.0.0 in the actions group ↗ huggingface/sentence-transformers

Quick answers

What shipped in Hugging Face on August 21, 2026?: 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. In total, 16 commits and 17 pull requests landed.
Who contributed to Hugging Face on August 21, 2026?: 6 developers shipped this update, including dependabot, jiqing-feng, Yih-Dar, Matt, YangKai0616, and danieldk.
What were the notable Hugging Face updates?: Port ESMC and ESMFold2 to Transformers (#46419), Always tie embeddings for LongT5 and Pop2Piano (#47620), and Retry transient network errors (RemoteDisconnected) in github_utils (#48124).

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 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

@huggingface Aug 18

Harbor-HF Walls Off Sandboxes While Serge Patrols Its Gates

Overnight, Hugging Face's internal infrastructure hardened its edge: Harbor-HF locks down sandbox capabilities, Serge's verify workflow stops dispatching with empty inputs, and dataset-viewer workers finally log their own startups.

python 76 shipped

Elsewhere on the wire

AI Agents Aug 22

Cline hubs get drain and replay as Langchain plumb gates

Cline just shipped rollover-grade hub commands while Langchain exposes gateway internals, and OpenHands quietly broke its own monitoring contract.

ai-agents 42 shipped

Local LLMs Aug 22

SGLang ships v0.5.18 with 710 PRs and DeepSeek V4 FP4 auto-routing

The biggest release of the week just landed: SGLang v0.5.18 adds new models, fixes a silent disagg bug, and defaults DeepSeek V4 FP4 checkpoints to the right MoE runner.