Who contributed to Hugging Face on May 19, 2026?

6 developers shipped this update, including kaixuanliu, pkooij, s1lent4gnt, jiqing-feng, sergiopaniego, and lhoestq.

What were the notable Hugging Face updates?

Fix remaining RAG doc examples that crash on current transformers (#46044), [CB] Remove OpenTelemetry (#45984), and fix model parallel issues for deimv2.

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

TRANSFORMERS FIXES RAG PIPELINE CRASHES; LEROBOT SHIPS SEEED ROBOT SUPPORT

By RepoJournal · Filed 06:04 UTC on May 19, 2026 · About Hugging Face

6 people shipped this

sergiopaniego @sergiopaniego 2 cited

kaixuanliu @kaixuanliu 1 cited

pkooij @pkooij 1 cited

s1lent4gnt @s1lent4gnt 1 cited

jiqing-feng @jiqing-feng 1 cited

lhoestq @lhoestq 1 cited

Transformers patched three critical bugs breaking RAG documentation examples and generation, while LeRobot added native Seeed Studio reBot arm integration that replaces fragile plugin packages.

The transformers team landed three urgent fixes across the RAG stack. First, documentation examples that crash on current transformers are now corrected [1], replacing deprecated checkpoints and removed API calls that broke the RAG retriever pipeline. Second, a missing AttributeError guard in RAG generate() [4] prevents crashes when RagConfig lacks num_return_sequences and num_beams fields. Third, OpenTelemetry instrumentation was removed [2] to simplify the codebase, and deimv2 model parallelism test failures are now fixed [3]. On the LeRobot side, Seeed Studio's reBot B601-DM arm is now a first-class citizen [5], eliminating the need for fragile out-of-tree plugin packages and providing both single-arm and bimanual follower configurations. Separately, LeRobot's reward configs now allow zero-shot models to skip optimizer presets [6], and accelerate fixed tied weight handling in quantized models [7] where lm_head wasn't staying in fp32. Repo2RLEnv made the LLM flag optional for mining pipelines [8] and corrected bootstrap test command generation [9]. Documentation fixes landed across accelerate [10] and lerobot [11].

Action items

→ Upgrade transformers immediately if using RAG examples or RagSequenceForGeneration in production huggingface/transformers [immediate]
→ If quantizing models with tied weights, pull the accelerate quantization fix for fp32 module preservation huggingface/accelerate [immediate]
→ LeRobot users: migrate from lerobot-robot-seeed-b601 plugins to native reBot B601-DM support huggingface/lerobot [plan]
→ Monitor Repo2RLEnv test command generation in your synthesis pipelines after the bootstrap fix huggingface/Repo2RLEnv [monitor]

References

[1] Fix remaining RAG doc examples that crash on current transformers (#46044) huggingface/transformers
[2] [CB] Remove OpenTelemetry (#45984) huggingface/transformers
[3] fix model parallel issues for deimv2 ↗ huggingface/transformers
[4] Fix AttributeError in RAG generate() for missing config fields (#46035) huggingface/transformers
[5] feat(robots): natively integrate Seeed Studio reBot B601-DM arm ↗ huggingface/lerobot
[6] Allow zero-shot reward configs without optimizer presets ↗ huggingface/lerobot
[7] Fix `keep_in_fp32_modules` not working for tied weights in `load_and_quantize_model` ↗ huggingface/accelerate
[8] make llm flag optional for mining pipelines ↗ huggingface/Repo2RLEnv
[9] fix: bootstrap saves executable test_cmds; normalize strips -q ↗ huggingface/Repo2RLEnv
[10] docs: Fix docstring of `fsdp2_prepare_auto_wrap_policy` (#4037) huggingface/accelerate
[11] Mention the new Lance LeRobotDataset implementation in the docs ↗ huggingface/lerobot

Quick answers

What shipped in Hugging Face on May 19, 2026?: Transformers patched three critical bugs breaking RAG documentation examples and generation, while LeRobot added native Seeed Studio reBot arm integration that replaces fragile plugin packages. In total, 27 commits and 27 pull requests landed.
Who contributed to Hugging Face on May 19, 2026?: 6 developers shipped this update, including kaixuanliu, pkooij, s1lent4gnt, jiqing-feng, sergiopaniego, and lhoestq.
What were the notable Hugging Face updates?: Fix remaining RAG doc examples that crash on current transformers (#46044), [CB] Remove OpenTelemetry (#45984), and fix model parallel issues for deimv2.

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.