Who contributed to Hugging Face on May 16, 2026?

4 developers shipped this update, including ivarflakstad, HaomingSong, assafvayner, and HuggingFaceInfra.

What were the notable Hugging Face updates?

[v1.15.0] Region-aware buckets & repos, `hf skills list`, polished CLI help and more, [Metal] Improved inter-encoder sync and gemv, and Fix transformers min version for tiny gemma4 as 5.5.0 (#5763).

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

HUB SHIPS REGION-AWARE BUCKETS, CANDLE UNLOCKS METAL GPU PARALLELISM

By RepoJournal · Filed 06:04 UTC on May 16, 2026 · About Hugging Face

4 people shipped this

ivarflakstad @ivarflakstad 1 cited

HaomingSong @HaomingSong 1 cited

assafvayner @assafvayner 1 cited

HuggingFaceInfra @HuggingFaceInfra 1 cited

Hugging Face Hub v1.15.0 lands region pinning for buckets and repos, while Candle's Metal backend jumps to inter-encoder parallelism with dependency-tracked synchronization.

The Hub's new region-aware API [1] lets you anchor buckets and repos to US or EU infrastructure at creation time, exposed via `create_bucket(region="us")` and CLI flags on both `hf buckets create` and `hf repos create`. This ships alongside a polished CLI help system and a new `hf skills list` command [1]. Over in Candle, the Metal backend [2] moves from semaphore-based encoder locking to mutex guards and dependency-tracked synchronization, enabling true inter-encoder parallelism when compute tasks have no dependencies. The gemv kernels now align with MLX and Candle Core specs [2]. TRL continues hardening: Gemma4 now requires transformers 5.5.0 minimum [3], and CI memory leaks are fixed by clearing chained exception tracebacks [4]. LeRobot's PI0.5 model is catching up to transformers 5.4.0's new embedding scaling internals, removing manual PaliGemma scaling that the framework now handles [5]. The Hub's FutureWarning for HF_HUB_ENABLE_HF_TRANSFER is now visible to end users, not buried under Python's default DeprecationWarning filter [8]. Documentation across the org is staying current: xet-core's repo reorganization forced Hub docs updates [6], and Inference Providers docs auto-regenerated after package bumps [7].

Action items

→ Update to huggingface_hub v1.15.0 to use region pinning on new buckets and repos huggingface/huggingface_hub [plan]
→ Patch transformers min version to 5.5.0 if you're testing Gemma4 in TRL pipelines huggingface/trl [monitor]
→ Review HF_HUB_ENABLE_HF_TRANSFER deprecation warnings; XET is the replacement path huggingface/huggingface_hub [plan]
→ Pull latest Candle Metal if you're benchmarking GPU kernels; inter-encoder parallelism unlocks new performance huggingface/candle [monitor]

References

[1] [v1.15.0] Region-aware buckets & repos, `hf skills list`, polished CLI help and more ↗ huggingface/huggingface_hub
[2] [Metal] Improved inter-encoder sync and gemv ↗ huggingface/candle
[3] Fix transformers min version for tiny gemma4 as 5.5.0 (#5763) huggingface/trl
[4] Fix OOM in CI by clearing chained exception tracebacks (#5776) huggingface/trl
[5] fix(pi05): update pi05 with transformers v5.4.0 scaled PaliGemma embeddings ↗ huggingface/lerobot
[6] docs(xet): update broken xet-core links after repo re-org ↗ huggingface/hub-docs
[7] [Bot] Update Inference Providers documentation ↗ huggingface/hub-docs
[8] Make HF_HUB_ENABLE_HF_TRANSFER deprecation warning visible to users (#4220) huggingface/huggingface_hub

Quick answers

What shipped in Hugging Face on May 16, 2026?: Hugging Face Hub v1.15.0 lands region pinning for buckets and repos, while Candle's Metal backend jumps to inter-encoder parallelism with dependency-tracked synchronization. In total, 15 commits, 13 pull requests, and 1 releases landed.
Who contributed to Hugging Face on May 16, 2026?: 4 developers shipped this update, including ivarflakstad, HaomingSong, assafvayner, and HuggingFaceInfra.
What were the notable Hugging Face updates?: [v1.15.0] Region-aware buckets & repos, `hf skills list`, polished CLI help and more, [Metal] Improved inter-encoder sync and gemv, and Fix transformers min version for tiny gemma4 as 5.5.0 (#5763).

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 9 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 9 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.