Who contributed to Hugging Face on July 28, 2026?

5 developers shipped this update, including linoytsaban, sayakpaul, Raushan Turganbay, Ferdinand Mom, and osolmaz.

What were the notable Hugging Face updates?

[lora training] add caption dropout and aspect ratio buckets to krea2 and qwen-image, [tests] migrate training tests to pytest, and Add FSDP plans to all models (#47165).

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

DIFFUSERS TIGHTENS LORA TRAINING, TRANSFORMERS SHIPS FSDP2 FOUNDATION

By RepoJournal · Filed 06:59 UTC on July 28, 2026 · About Hugging Face

5 people shipped this

Ferdinand Mom @3outeille 2 cited

osolmaz @osolmaz 2 cited

linoytsaban @linoytsaban 1 cited

sayakpaul @sayakpaul 1 cited

Raushan Turganbay @zucchini-nlp 1 cited

Hugging Face cut training tests over to pytest while shipping caption dropout and aspect-ratio bucketing for two key LoRA models, and transformers landed its FSDP2 native module ahead of major distributed training rework.

Diffusers merged caption dropout and aspect-ratio bucketing for krea2 and qwen-image [1], two training features that let operators randomly zero captions to improve classifier-free guidance and pack images by ratio instead of padding to fixed sizes. The team also migrated training tests to pytest [2] to standardize infrastructure, with follow-up work to consolidate test mixins and eliminate copy-paste patterns [2]. Transformers shipped FSDP2 native module wrapping and plan verification [3], moving distributed training coordination out of integrations and into core, then cleaned up redundant guards across the codebase [4]. On the multimodal front, transformers updated most commonly used LLM processors in a breaking change that removed small private functions [5]. MLClaw completed OpenAI OAuth integration across five commits, removing the Codex CLI dependency entirely and routing agent models through native OpenClaw harness [6], while hardening credential lifecycle to prevent logout undo and preserve migration intent [7].

Action items

→ Review breaking processor changes in transformers [ref:9] before upgrading if you depend on private multimodal functions huggingface/transformers [plan]
→ Test FSDP2 integration if running distributed training at scale huggingface/transformers [monitor]
→ Integrate caption dropout into LoRA training pipelines for better CFG performance huggingface/diffusers [plan]

References

[1] [lora training] add caption dropout and aspect ratio buckets to krea2 and qwen-image ↗ huggingface/diffusers
[2] [tests] migrate training tests to pytest. ↗ huggingface/diffusers
[3] Add FSDP plans to all models (#47165) huggingface/transformers
[4] Remove redundant guarding for distributed (#47570) huggingface/transformers
[5] :rotating_light: Processors update the rest ↗ huggingface/transformers
[6] fix(openai): use Codex agent runtime ↗ huggingface/mlclaw
[7] fix(credentials): harden Codex credential lifecycle ↗ huggingface/mlclaw

Quick answers

What shipped in Hugging Face on July 28, 2026?: Hugging Face cut training tests over to pytest while shipping caption dropout and aspect-ratio bucketing for two key LoRA models, and transformers landed its FSDP2 native module ahead of major distributed training rework. In total, 25 commits, 26 pull requests, and 1 releases landed.
Who contributed to Hugging Face on July 28, 2026?: 5 developers shipped this update, including linoytsaban, sayakpaul, Raushan Turganbay, Ferdinand Mom, and osolmaz.
What were the notable Hugging Face updates?: [lora training] add caption dropout and aspect ratio buckets to krea2 and qwen-image, [tests] migrate training tests to pytest, and Add FSDP plans to all models (#47165).

TRANSFORMERS HARDENS CI RELAY; HUB FIXES XET CACHE BUG

Serge review automation is now reachable behind the VPN, and a critical xet file download issue tied to redacted hashes in gated repos is resolved.

python 35 shipped 2-min read

@huggingface 2 days ago

TRANSFORMERS-CI SHIPS INTELLIGENT TEST TRIAGE, CUTS LLM NOISE 100X

Hugging Face's serge verify loop now distills 100KB failure tracebacks down to the actual assertion, runs every test 5 times to catch flakiness, and dispatches each failure category with a custom instruction before spinning GPU or consuming a single LLM token.

python 45 shipped 2-min read

@huggingface 3 days ago

TRL CI BURNS, BITSANDBYTES PINNED BELOW 0.50.0

Hugging Face's TRL training library is firefighting two separate CI failures overnight, forcing temporary pins on bitsandbytes and xfailing critical tests against a broken transformers dev build.

python 7 shipped 1-min read

@huggingface 4 days ago

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

Elsewhere on the wire

AI Agents about 3 hours ago

LANGCHAIN CORES SHIP, ANTHROPIC THINKING BLOCKS FIXED, OPENHANDS TIGHTENS WEBSOCKET AUTH

LangChain released two core packages with critical fixes for streaming stability and environment variable handling, while OpenHands closed a security gap that left WebSocket connections unprotected.

ai-agents 119 shipped 1-min read

Local LLMs about 3 hours ago

OLLAMA RIPS OUT IMAGE GENERATION, LLAMA.CPP SHIPS SPECULATIVE DECODING

Ollama removed its entire MLX image generation pipeline overnight, while llama.cpp landed DSpark speculative decoding and fixed a critical SYCL memory bug affecting Arc GPUs.

+11

llms 239 shipped 2-min read

@CachyOS about 3 hours ago

KERNEL SCHEDULER UPDATES ACROSS BORE AND SCX TOOLING

CachyOS shipped BORE 6.8.0 patches across two kernel series while upgrading its sched-ext ecosystem in parallel, signaling major tuning work ahead.

infra 83 shipped 1-min read

Elixir & Phoenix about 3 hours ago

ELIXIR TIGHTENS ERROR HANDLING ACROSS DATE, ECTO, AND IEXHELPERS

Date.range now surfaces ArgumentError instead of Protocol.UndefinedError when passed invalid step values, catching a class of bugs that silently fail in production code.