RepoJournal
Hugging Face

@huggingface

Transformers, Datasets, and the open AI-model layer

Keep up with Hugging Face in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

One email a day. Unsubscribe in one click.

Pick a date

  1. Sun 28 jun
  2. Mon 29 jun
  3. Tue 30 jun
  4. Wed 1 jul
  5. Thu 2 jul
  6. Fri 3 jul
  7. Sat 4 jul
  8. Sun 5 jul
  9. Mon 6 jul
  10. Tue 7 jul
  11. Wed 8 jul
  12. Thu 9 jul
  13. Fri 10 jul
  14. Sat 11 jul
  15. Sun 12 jul
  16. Mon 13 jul
  17. Tue 14 jul
  18. Wed 15 jul
  19. Thu 16 jul
  20. Fri 17 jul
  21. Sat 18 jul
  22. Sun 19 jul
  23. Mon 20 jul
  24. Tue 21 jul
  25. Wed 22 jul
  26. Thu 23 jul
  27. Fri 24 jul
  28. Sat 25 jul
  29. Sun 26 jul
  30. Mon 27 jul
Topics: Python AI / ML Full archive →

The Wire · Showcase

DISTILLATION TRAINER GRADIENT ACCUMULATION BUG FIXED, LEROBOT IMAGE STATS OVERFLOW PATCHED

By RepoJournal · Filed · About Hugging Face

dependabot[bot] behroozazarkhalili Altman-conquer qgallouedec
4 people shipped this
dependabot[bot] @dependabot[bot] 2 cited behroozazarkhalili @behroozazarkhalili 1 cited Altman-conquer @Altman-conquer 1 cited qgallouedec @qgallouedec 1 cited

TRL's distillation trainers were silently miscalculating loss under gradient accumulation, and LeRobot's image statistics were overflowing to zero for valid data.

The GKDTrainer, GOLDTrainer, and DistillationTrainer accepted `num_items_in_batch` in their loss computation but never used it [1], causing JSD distillation loss to normalize by local microbatch token count instead of global count. Under gradient accumulation, this breaks the gradient scaling that transformers' base Trainer expects, silently producing wrong gradients. This is a critical fix for anyone training distilled models at scale [2]. In parallel, LeRobot's image statistics computation was promoting uint8 samples to float *after* squaring them in RunningQuantileStats, causing uint8 overflow that made computed variance negative and clamped to zero [3], so stats.json reported `std=0` for non-constant image data [4]. Both fixes are merged and ready. TRL also patched the GLM-4-MoE chat template to properly terminate assistant turns with role markers instead of missing end-of-turn tokens [5]. Routine dependency bumps across trl and ml-intern [6], [7]. Chat-UI increased MiniMax-M3's max_tokens to 65536 to prevent the router from truncating reasoning-heavy outputs at 2048 tokens [8].

One email a day. Unsubscribe in one click.

Action items

References

  1. [1] Normalize JSD distillation loss by num_items_in_batch for gradient accumulation ↗ huggingface/trl
  2. [2] Normalize JSD distillation loss by num_items_in_batch for gradient accumulation (#6006) huggingface/trl
  3. [3] fix(datasets): avoid uint8 overflow in image stats (#3697) huggingface/lerobot
  4. [4] fix(datasets): avoid uint8 overflow in image stats ↗ huggingface/lerobot
  5. [5] [fix] GLM-4-MoE template: turn-terminating token to the turn itself ↗ huggingface/trl
  6. [6] Bump the actions group with 4 updates ↗ huggingface/trl
  7. [7] Bump the actions group with 4 updates ↗ huggingface/ml-intern
  8. [8] Set max_tokens 65536 for MiniMax-M3 huggingface/chat-ui

Quick answers

What shipped in Hugging Face on June 14, 2026?
TRL's distillation trainers were silently miscalculating loss under gradient accumulation, and LeRobot's image statistics were overflowing to zero for valid data. In total, 5 commits and 6 pull requests landed.
Who contributed to Hugging Face on June 14, 2026?
4 developers shipped this update, including behroozazarkhalili, Altman-conquer, qgallouedec, and dependabot[bot].
What were the notable Hugging Face updates?
Normalize JSD distillation loss by num_items_in_batch for gradient accumulation, Normalize JSD distillation loss by num_items_in_batch for gradient accumulation (#6006), and fix(datasets): avoid uint8 overflow in image stats (#3697).

More from @huggingface

Hugging Face @huggingface

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.

Ferdinand Mom tarekziade evalstate alejandro-ao +1
python 45 shipped 2-min read
Hugging Face @huggingface

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.

albertvillanova rwightman
python 7 shipped 1-min read
Hugging Face @huggingface

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.

Matt Cyril Vallez Anton Vlasjuk qgallouedec +5
python 70 shipped 2-min read
Hugging Face @huggingface

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.

evalstate Ferdinand Mom Kashif Rasul Harry Mellor +10
python 61 shipped 1-min read

Daily updates, in your inbox

Follow Hugging Face

Keep up with Hugging Face in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

One email a day. Unsubscribe in one click. Read a past issue →

Elsewhere on the wire

AI Agents AI Agents

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.

langchain-oss-model-profiles langchain-oss-model-profiles[bot] saoudrizwan
ai-agents 16 shipped 2-min read
Local LLMs Local LLMs

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.

github-actions[bot] dhiltgen timkhronos Pascal +10
llms 125 shipped 1-min read
CachyOS @CachyOS

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.

Piotr Gorski github-actions Vladislav Nepogodin ptr1337 +7
infra 97 shipped 1-min read
Elixir & Phoenix Elixir & Phoenix

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.

Benjamin Milde Alexander Gubarev lukaszsamson makoto-developer
elixir 10 shipped 1-min read
Want every project, not just this one?

We use privacy-friendly analytics (Google Analytics, IP-anonymized) to see which pages help readers. No ads, and we never sell your data. See our Privacy Policy.