Who contributed to Hugging Face on July 14, 2026?

8 developers shipped this update, including CarolinePascal, Cyril Vallez, Sunt-ing, Sreekant Baheti, drbh, Albert Villanova del Moral, Sergio Paniego Blanco, and Quentin Gallouédec.

What were the notable Hugging Face updates?

[generate] Add proper MTP support (#46229), Fix DPO/KTO `use_liger_kernel` under DeepSpeed ZeRO-3, and feat(video): viridis colormap for grayscale/depth feeds.

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

TRANSFORMERS ADDS MTP DECODING, TRL FIXES DEEPSEEP ZERO-3 CRASH

By RepoJournal · Filed 06:59 UTC on July 14, 2026 · About Hugging Face

8 people shipped this

Quentin Gallouédec @qgallouedec 2 cited

CarolinePascal @CarolinePascal 1 cited

Cyril Vallez @Cyrilvallez 1 cited

Sunt-ing @Sunt-ing 1 cited

Sreekant Baheti @Sreekant13 1 cited

drbh @drbh

Albert Villanova del Moral @albertvillanova

Sergio Paniego Blanco @sergiopaniego

Transformers finally ships proper multi-token prediction support [ref:3], while TRL patches a critical DeepSpeed ZeRO-3 incompatibility that breaks Liger kernel fusion in production training.

In brief

Transformers now decodes MTP (Mixture of Tokens and Probabilities) properly. If you've been getting garbled outputs from models using MTP, this unblocks you. Worth a quick upgrade if that's your setup.
TRL patched DPO/KTO under DeepSpeed ZeRO-3 with `use_liger_kernel` enabled. That crash? Fixed. Means you can train alignment models at scale again without ritual sacrifices to the memory gods.
Qwen Omni's batched text postprocessing got a fix in Transformers. Batching shouldn't fail silently—now it doesn't. Small fix, big quality-of-life win for inference pipelines.
Transformers' `get_json_schema` no longer crashes when docstring choices aren't strings. One of those "how did this ever work" bugs that probably lived in production for months.
TRL now drops fully-masked examples after truncation in SFT. Cleaner training data. Seems simple but prevents dead weights on useless tokens during fine-tuning.

MTP decoding lands in Transformers [1], letting models that support multi-token prediction generate sequences faster without API changes. This is the feature teams have been waiting for in speculative decoding pipelines. Meanwhile, TRL's Liger kernel optimization was silently breaking under DeepSpeed ZeRO-3 [2] because the fused DPO loss tries to read `lm_head.weight` directly instead of through the model's forward hook, where ZeRO-3 re-materializes sharded parameters. That fix mirrors a similar GRPO bug from weeks ago. On the visualization front, LeRobot's dataset viewer now renders grayscale and depth camera feeds with the viridis colormap [3], making it actually possible to inspect robot vision data without custom scripts. Three smaller fixes ship across the stack: Qwen Omni's batched text postprocessing now handles full batches [4], `get_json_schema` stops crashing on non-string enum choices [5], and SFT training now drops fully-masked examples after truncation [6] to prevent silent training bugs.

Action items

→ Review MTP support in Transformers if you use speculative decoding or multi-token prediction models huggingface/transformers [plan]
→ Patch DPO/KTO training immediately if running with use_liger_kernel=True and DeepSpeed ZeRO-3 huggingface/trl [immediate]
→ Update TRL to get SFT truncation fix and avoid silently truncated training data huggingface/trl [plan]
→ Pull Qwen Omni fix if you have batched text-only pipelines returning truncated outputs huggingface/transformers [plan]

References

[1] [generate] Add proper MTP support (#46229) huggingface/transformers
[2] Fix DPO/KTO `use_liger_kernel` under DeepSpeed ZeRO-3 ↗ huggingface/trl
[3] feat(video): viridis colormap for grayscale/depth feeds ↗ huggingface/lerobot-dataset-visualizer
[4] Fix Qwen Omni batched text postprocessing ↗ huggingface/transformers
[5] Fix get_json_schema crash on non-string docstring choices (#47072) huggingface/transformers
[6] Drop fully-masked examples after truncation in SFT (#6320) huggingface/trl

Quick answers

What shipped in Hugging Face on July 14, 2026?: Transformers finally ships proper multi-token prediction support [ref:3], while TRL patches a critical DeepSpeed ZeRO-3 incompatibility that breaks Liger kernel fusion in production training. In total, 23 commits and 23 pull requests landed.
Who contributed to Hugging Face on July 14, 2026?: 8 developers shipped this update, including CarolinePascal, Cyril Vallez, Sunt-ing, Sreekant Baheti, drbh, Albert Villanova del Moral, Sergio Paniego Blanco, and Quentin Gallouédec.
What were the notable Hugging Face updates?: [generate] Add proper MTP support (#46229), Fix DPO/KTO `use_liger_kernel` under DeepSpeed ZeRO-3, and feat(video): viridis colormap for grayscale/depth feeds.

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 2 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 3 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

@huggingface 4 days 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

Elsewhere on the wire

AI Agents about 21 hours ago

CLINE SHIPS FIVE CRITICAL AGENT FIXES AFTER END-TO-END TESTING BLITZ

Saoudrizwan just landed a cascade of fixes that prevent the agent loop from running wild, stop file reads from resolving to the wrong directory, and plug the compaction pipeline.

ai-agents 16 shipped 2-min read

Local LLMs about 21 hours ago

LLAMA.CPP SHIPS MINIMAX-M3 SPARSE ATTENTION, VLLM UNLOCKS HETEROGENEOUS PREFILL-DECODE

The inference stack gets a new model architecture and a fundamental shift in how prefill and decode can run on different hardware configurations.

+10

llms 125 shipped 1-min read

@CachyOS about 21 hours ago

GLIBC 2.43 REBUILD SHIPS, INSTALLER JUMPS TO C++26

CachyOS pushed glibc 2.43 through the pipeline while the CLI installer upgraded its C++ standard to C++26, signaling a major refresh across core infrastructure.

infra 97 shipped 1-min read

Elixir & Phoenix about 21 hours ago

ECTO FIXES SUBQUERY PLANNING IN JOINS, ELIXIR PATCHES IEX PARSING

Ecto's query planner now handles subqueries correctly in interpolated join filters, while Elixir fixes a syntax error that broke 'not in' as a continuation operator in IEx.