Who contributed to Hugging Face on August 19, 2026?

12 developers shipped this update, including jiqing-feng, danieldk, drbh, Yih-Dar, Priyans-Lathiya, David Quinlan, AmineDiro, and albertvillanova, and 4 more.

What were the notable Hugging Face updates?

Crescent Island (cri) support for the XPU/SYCL build, Add locking data structures to `kernels-data`, and kernel-builder: move `build.toml` parsing to `kernels-data`.

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

TWO NEW TOOLCHAINS AND A PATCH THAT MATTERS

By RepoJournal · Filed 06:59 UTC on August 19, 2026 · About Hugging Face

12 people shipped this

danieldk @danieldk 3 cited

jiqing-feng @jiqing-feng 1 cited

David Quinlan @DavidJohnQuinlan 1 cited

albertvillanova @albertvillanova 1 cited

YiYi Xu @yiyixuxu 1 cited

Sayak Paul @sayakpaul 1 cited

drbh @drbh

Yih-Dar @ydshieh

Priyans-Lathiya @Priyans-Lathiya

AmineDiro @AmineDiro

Dhruv Nair @DN6

Buffett Liu @buffett0323

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.

Intel's Crescent Island support lands in huggingface/kernels [1], requiring a significant toolchain bump: ocloc jumps to 26.22.38646.4 and sycl-tla to 0.9.2, with AOT gated on DPC++ 2026.0. If you're building XPU/SYCL kernels, this is the upgrade you've been waiting for, but only if you're ready for the new toolchain requirements. Meanwhile, danieldk's trio of PRs prepare kernels-data for dependency management [2], moving build.toml parsing into kernels-data [3], and adding a Hub fetcher [4]. This groundwork matters because it signals a shift toward proper kernel dependencies, but today it's mostly internal plumbing. In transformers, a critical bug fix ensures encoder_attention_mask is no longer silently discarded in GPT2 cross-attention [5], which could affect any encoder-decoder task you're running. The release notes say, "the new attn mask interface (#42848) rewired GPT2Model... but added an encoder_attention_mask = None line directly above the call that overwrote the incoming mask before it was passed in." If you've seen degraded summarization or translation outputs, this might be why. On the diffusers side, Modular Diffusers is no longer experimental [6], and a torchao quantization fix routes root-level tensors around safetensors reconstruction [7]. Finally, trl restores Liger Kernel's global patches between tests to prevent cross-test contamination [8], a quiet but important fix for anyone running the test suite with use_liger_kernel enabled. With 42 commits and 42 PRs across the stack, nothing else demands immediate attention.

Action items

→ Review your GPT2-based encoder-decoder pipelines for mask handling after upgrading transformers huggingface/transformers [immediate]
→ If you build XPU/SYCL kernels, plan for the toolchain bump to support Crescent Island huggingface/kernels [plan]
→ Watch the kernels-data dependency groundwork; it will enable kernel-kernel dependencies soon huggingface/kernels [monitor]

References

[1] Crescent Island (cri) support for the XPU/SYCL build ↗ huggingface/kernels
[2] Add locking data structures to `kernels-data` ↗ huggingface/kernels
[3] kernel-builder: move `build.toml` parsing to `kernels-data` ↗ huggingface/kernels
[4] `fetchFromHuggingFace`: add ↗ huggingface/kernels
[5] [GPT2] Fix encoder_attention_mask being silently discarded in cross-attention (#47946) huggingface/transformers
[6] Drop the experimental warning from Modular Diffusers (#14525) huggingface/diffusers
[7] [tests] refactor pipeline-level quantization tests (#14435) huggingface/diffusers
[8] Restore Liger Kernel's global patches between tests ↗ huggingface/trl

Quick answers

What shipped in Hugging Face on August 19, 2026?: 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. In total, 42 commits and 42 pull requests landed.
Who contributed to Hugging Face on August 19, 2026?: 12 developers shipped this update, including jiqing-feng, danieldk, drbh, Yih-Dar, Priyans-Lathiya, David Quinlan, AmineDiro, and albertvillanova, and 4 more.
What were the notable Hugging Face updates?: Crescent Island (cri) support for the XPU/SYCL build, Add locking data structures to `kernels-data`, and kernel-builder: move `build.toml` parsing to `kernels-data`.

TRL DROPS ASYNC DISTILLATION, DIFFUSERS GO TENSOR PARALLEL

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

python 22 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

@huggingface Aug 17

HARBOR-HF SHIPS UNIFIED TYPESCRIPT CONTROL SERVICE, TRANSFORMERS PATCHES DEVICE MISMATCHES

Harbor-HF migrated from distributed Python services to a single hosted TypeScript control plane, while Transformers fixed critical device placement bugs and unlocked tokenizers 0.23.1 compatibility.

python 22 shipped

@huggingface Aug 16

AGENT-MANAGER FIXES DRAG-AND-DROP, MOBILE LAYOUT, AND TAB REFRESH

The sidebar drag-and-drop broke when groups became framed UI elements, and a series of layout and polling fixes landed to smooth mobile and desktop rendering.

python 25 shipped

Elsewhere on the wire

AI Agents Aug 20

LangChain ships core 1.6.0 with a security warning tucked inside

LangChain's newest core release fixes a tool-injection bug, but the fix itself widens the path for arbitrary code execution.

ai-agents 95 shipped

Local LLMs Aug 20

OLLAMA FIXES DEADLOCK THAT HUNG CHAT AND GENERATE

A mid-stream parser error in Ollama's server silently wedged chat and generate requests, leaking goroutines and hanging retries until a fix landed overnight.

+14

llms 268 shipped

@CachyOS Aug 20

LINUX-CACHYOS SHIPS 7.2.0-1

CachyOS's custom kernel just advanced to 7.2.0-1, and the sched-ext toolchain is moving in lockstep.

infra 55 shipped

Elixir & Phoenix Aug 20

LIVE VIEW HOOKS GET THE ELOQUENT TREATMENT

Phoenix Live View is shipping a passel of hook and JS-client fixes today, headlined by new beforeUpdate arguments that hand you more control over a hook's lifespan.