Who contributed to Hugging Face on May 23, 2026?

4 developers shipped this update, including HaomingSong, coyotte508, Enderfga, and XciD.

What were the notable Hugging Face updates?

fix(pi0, pi05): stabilize torch.compile and expand test coverage, fix(deps): cap placo below 0.9.16 and harden kinematics import (#3647), and hub-v2.13.0.

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

LEROBOT STABILIZES TORCH.COMPILE, HUB.JS SHIPS STREAMING APIS

By RepoJournal · Filed 06:04 UTC on May 23, 2026 · About Hugging Face

4 people shipped this

HaomingSong @HaomingSong 1 cited

coyotte508 @coyotte508 1 cited

Enderfga @Enderfga 1 cited

XciD @XciD 1 cited

LeRobot fixes critical graph breaks in PI0/PI0.5 torch.compile while hub.js lands async-generator download progress tracking and safer tensor type sync.

LeRobot's torch.compile pipeline got two critical fixes [1]: deepcopy of DynamicCache no longer breaks the computation graph in sample_actions, and repeated Dynamo recompilation from compute_layer_complete is eliminated. New compile tests and comprehensive end-to-end coverage [1] ensure these stay fixed. Separately, a dependency cap on placo <0.9.16 [2] patches nightly test failures on Ubuntu 24.04 where liburdfdom_sensor.so.4 goes missing. Hub.js v2.13.0 [3] ships streaming progress events for copyFolder [4] via async generators, letting you track downloads of many small files not tracked by git; this is medium risk since it refactors control flow around commits. The same release syncs safetensors dtypes with upstream [5], adding C64, F8_E4M3FNUZ, F8_E5M2FNUZ, U64, and U32. Diffusers adds AnyFlow any-step video pipelines [6], a flow-map framework that scales monotonically with NFE (1 to 32 steps) from a single checkpoint. And huggingface_hub migrates PyPI publishing to Trusted Publishing with Sigstore attestations [7], closing the supply-chain gap that left recent releases unsigned on PyPI.

Action items

→ Review and merge LeRobot torch.compile fixes [ref:1] before next release cycle huggingface/lerobot [plan]
→ Pin placo <0.9.16 in dependency lockfiles if deploying on Ubuntu 24.04 [ref:2] huggingface/lerobot [immediate]
→ Test hub.js v2.13.0 streaming APIs in non-git-tracked file workflows [ref:6][ref:7] huggingface/huggingface.js [plan]

References

[1] fix(pi0, pi05): stabilize torch.compile and expand test coverage ↗ huggingface/lerobot
[2] fix(deps): cap placo below 0.9.16 and harden kinematics import (#3647) huggingface/lerobot
[3] hub-v2.13.0 ↗ huggingface/huggingface.js
[4] Streaming events for copyFolder ↗ huggingface/huggingface.js
[5] [hub] Sync safetensors Dtype union with upstream (#2187) huggingface/huggingface.js
[6] Add AnyFlow Any-Step Video Diffusion Pipelines (Bidirectional + FAR Causal) ↗ huggingface/diffusers
[7] ci(release): migrate PyPI publish to Trusted Publishing + attestations ↗ huggingface/huggingface_hub

Quick answers

What shipped in Hugging Face on May 23, 2026?: LeRobot fixes critical graph breaks in PI0/PI0.5 torch.compile while hub.js lands async-generator download progress tracking and safer tensor type sync. In total, 17 commits, 16 pull requests, and 1 releases landed.
Who contributed to Hugging Face on May 23, 2026?: 4 developers shipped this update, including HaomingSong, coyotte508, Enderfga, and XciD.
What were the notable Hugging Face updates?: fix(pi0, pi05): stabilize torch.compile and expand test coverage, fix(deps): cap placo below 0.9.16 and harden kinematics import (#3647), and hub-v2.13.0.

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.