Who contributed to Hugging Face on June 10, 2026?

5 developers shipped this update, including akshan-main, remi-or, michaelbenayoun, tarekziade, and jiqing-feng.

What were the notable Hugging Face updates?

refactor unet_2d tests, refactor unet_1d tests, and refactor unet tests (3d_condition, motion, controlnetxs).

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

DIFFUSERS COMPLETES UNET TEST MIGRATION, TRANSFORMERS SHIPS KERNEL FUSION

By RepoJournal · Filed 06:03 UTC on June 10, 2026 · About Hugging Face

5 people shipped this

akshan-main @akshan-main 3 cited

remi-or @remi-or 1 cited

michaelbenayoun @michaelbenayoun 1 cited

tarekziade @tarekziade 1 cited

jiqing-feng @jiqing-feng 1 cited

The diffusers team finished migrating all UNet test suites to a unified mixin-based structure, while transformers unlocked a cleaner kernel fusion API that puts control back in the hands of kernel authors.

Diffusers closed out a major internal refactor across five separate PRs. The UNet2D, UNet1D, and three variants of UNet3D test suites [1] [2] [3] moved from scattered test logic to a consistent Config + ModelTesterMixin pattern, with deprecated xformers and attention slicing tests dropped in favor of a cleaner test hierarchy. This standardization matters because it makes the test suite maintainable as the pipeline ecosystem grows. On the transformers side, two critical infrastructure improvements landed: the `max_requests_per_batch` parameter [4] [6] now separates token-sized tensors from request-sized ones in continuous batching, fixing a longstanding memory calculation problem, while a new KernelConfig API [5] [7] simplifies n-to-1 kernel fusion by making module fusion and parameter transformation explicit. The kernels repo gained CPU kernel skills [10] for the kernel-builder, adding a two-phase correctness-then-performance workflow with benchmarking and profiling built in. Serge shipped a critical security isolation layer [8] [9] that sandboxes fork PR code from the persistent review host, preventing tampered context scripts or imports from accessing GitHub App keys and LLM credentials.

Action items

→ Review the diffusers UNet test migration for any custom overrides in your fork huggingface/diffusers [plan]
→ Integrate max_requests_per_batch into continuous batching config if you own inference huggingface/transformers [plan]
→ Deploy serge security patch immediately if running PR review automation huggingface/serge [immediate]
→ Explore new KernelConfig fusion API for custom kernel development huggingface/kernels [monitor]

References

[1] refactor unet_2d tests ↗ huggingface/diffusers
[2] refactor unet_1d tests ↗ huggingface/diffusers
[3] refactor unet tests (3d_condition, motion, controlnetxs) ↗ huggingface/diffusers
[4] [CB] Add a "max_requests_per_batch" parameter ↗ huggingface/transformers
[5] Extended & simplified n-to-1 kernel fusion via KernelConfig ↗ huggingface/transformers
[6] [CB] Add a "max_requests_per_batch" parameter (#46434) huggingface/transformers
[7] Extended & simplified n-to-1 kernel fusion via KernelConfig (#46339) huggingface/transformers
[8] Isolate PR/fork code from the review host (#10) huggingface/serge
[9] Isolate PR/fork code from the review host ↗ huggingface/serge
[10] Add CPU kernel skills ↗ huggingface/kernels

Quick answers

What shipped in Hugging Face on June 10, 2026?: The diffusers team finished migrating all UNet test suites to a unified mixin-based structure, while transformers unlocked a cleaner kernel fusion API that puts control back in the hands of kernel authors. In total, 23 commits and 21 pull requests landed.
Who contributed to Hugging Face on June 10, 2026?: 5 developers shipped this update, including akshan-main, remi-or, michaelbenayoun, tarekziade, and jiqing-feng.
What were the notable Hugging Face updates?: refactor unet_2d tests, refactor unet_1d tests, and refactor unet tests (3d_condition, motion, controlnetxs).

TRANSFORMERS OVERHAULS LINEAR ATTENTION WHILE DEPRECATING LEGACY RESPONSE SCHEMA

The transformers library is retiring its fragile response_schema prototype in favor of streaming-compatible parsing, while simultaneously refactoring every linear attention model to use standardized convolution patterns.

python 70 shipped 2-min read

@huggingface 1 day ago

TRANSFORMERS SHIPS FSDP DISTRIBUTED TRAINING STACK, HUB LIBRARY PLUGS REDOS HOLE

Hugging Face landed distributed training orchestration in transformers while plugging a regex vulnerability that could stall untrusted card parsing for minutes.

+10

python 61 shipped 1-min read

@huggingface 2 days ago

TRANSFORMERS HARDENS AGAINST PYTORCH FRAGMENTATION WHILE TRL SIMPLIFIES DISTILLATION

Transformers plugged a cascading import failure that breaks downstream CI on older PyTorch versions, while TRL rips out dead code to lock DistillationTrainer into prompt-only datasets.

python 70 shipped 2-min read

@huggingface 3 days ago

DATASET VIEWER LOCKS DOWN ARROW, FUNES SHIPS GROUNDED ASK

Hugging Face security teams moved overnight to contain a critical Arrow IPC parsing vulnerability in dataset-viewer while shipping three production hardening releases across Repo2RLEnv, optimum-executorch, and funes.

python 91 shipped 2-min read

Elsewhere on the wire

AI Agents about 10 hours ago

CLAUDE OPUS 5 LANDS ACROSS THE STACK

The newest Anthropic model is now live in langchain, Cline, and llama-index, with native support for extended reasoning and 1M context windows.

ai-agents 28 shipped 1-min read

Local LLMs about 10 hours ago

OLLAMA LANDS LAGUNA SUPPORT AND CRUSHES MEMORY LEAKS WHILE SGLANG HITS V0.5.16 WITH CONFIDENCE-DRIVEN SPECULATIVE DECODING

Ollama shipped three critical performance and reliability fixes for Metal residency and concurrent access patterns, while SGL-Lang released 0.5.16 with a new speculative algorithm hitting 383.7 tok/s on DeepSeek-V4.