Who contributed to Hugging Face on August 6, 2026?

9 developers shipped this update, including andimarafioti, PigeonDan1, akshan-main, apolinario, mjun0812, Rémi Ouazan, Anton Vlasjuk, and vkuzo, and 1 more.

What were the notable Hugging Face updates?

v0.2.12, feat(vad): integrate Smart Turn v3.2 endpointing, and Merge pull request #429 from huggingface/agent/clean-dependency-extras.

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

SPEECH-TO-SPEECH 0.2.12 SHIPS WITH SMARTER TURN-TAKING AND WEBRTC SUPPORT

By RepoJournal · Filed 07:00 UTC on August 6, 2026 · About Hugging Face

9 people shipped this

andimarafioti @andimarafioti 3 cited

akshan-main @akshan-main 3 cited

Rémi Ouazan @remi-or 2 cited

apolinario @apolinario 1 cited

mjun0812 @mjun0812 1 cited

Anton Vlasjuk @vasqu 1 cited

vkuzo @vkuzo 1 cited

guarin @guarin 1 cited

PigeonDan1 @PigeonDan1

The final 0.2.x release lands Smart Turn v3.2 endpointing, WebRTC support, and direct audio input for audio-capable LLMs, while diffusers onboards MiniMax-H3 for joint video and audio generation.

Speech-to-speech 0.2.12 [1] closes the 0.2.x line with quantized CPU-based Smart Turn v3.2 endpointing enabled by default in realtime mode, distinguishing completed turns from mid-thought pauses without requiring GPU inference [2]. The release also ships WebRTC support, direct audio input for audio-capable LLMs, and a significantly improved browser demo alongside cleanup of stale dependency extras [3]. On the diffusers side, MiniMax-H3 [4] lands as a full modular implementation for joint video and audio generation, generating stereo video and synchronized audio in a single denoising pass with checkpoints housed on the Hub. Test infrastructure across diffusers continues its overhaul: Flux2 and QwenImage inpaint pipelines migrated to the new mixin structure [5] [6], while migrated tests now use `assert_tensors_close` for consistency [7]. Sentence-transformers fixed a critical DataLoader worker leak [8] that exhausted file descriptors when training with `dataloader_persistent_workers=True`, mirroring the protection transformers implemented. Transformers itself ships fixes for MLA cache compression [10], KimiK25 dimension swaps [9], linear attention kernel refactoring with native kernel fallbacks [11], torchao integration compatibility [12], and DAC offload test reliability [13].

Action items

→ Upgrade speech-to-speech to 0.2.12 if you use realtime mode with turn-taking huggingface/speech-to-speech [plan]
→ Apply sentence-transformers DataLoader worker fix immediately if using persistent_workers with dict eval_dataset huggingface/sentence-transformers [immediate]
→ Monitor transformers MLA and linear attention changes if you deploy these models in production huggingface/transformers [monitor]

References

[1] v0.2.12 ↗ huggingface/speech-to-speech
[2] feat(vad): integrate Smart Turn v3.2 endpointing ↗ huggingface/speech-to-speech
[3] Merge pull request #429 from huggingface/agent/clean-dependency-extras huggingface/speech-to-speech
[4] Add MiniMax-H3 ↗ huggingface/diffusers
[5] refactor flux2 pipeline tests to the new mixin structure ↗ huggingface/diffusers
[6] refactor qwenimage inpaint pipeline tests to the new mixin structure ↗ huggingface/diffusers
[7] use assert_tensors_close in the migrated pipeline tests ↗ huggingface/diffusers
[8] [`trainer`] Fix eval DataLoader worker leak with dataloader_persistent_workers ↗ huggingface/sentence-transformers
[9] [Fix] Swapped height and width in KimiK25 (#47786) huggingface/transformers
[10] [Fix] Fix multi-head latent attention (MLA) ↗ huggingface/transformers
[11] 🚨 [`Kernels`] Refactor all linear attn models & native kernels fallback (#47630) huggingface/transformers
[12] Migrate torchao integration off deleted torchao.dtypes ↗ huggingface/transformers
[13] Fix Dac offload tests (#47775) huggingface/transformers

Quick answers

What shipped in Hugging Face on August 6, 2026?: The final 0.2.x release lands Smart Turn v3.2 endpointing, WebRTC support, and direct audio input for audio-capable LLMs, while diffusers onboards MiniMax-H3 for joint video and audio generation. In total, 31 commits, 25 pull requests, and 1 releases landed.
Who contributed to Hugging Face on August 6, 2026?: 9 developers shipped this update, including andimarafioti, PigeonDan1, akshan-main, apolinario, mjun0812, Rémi Ouazan, Anton Vlasjuk, and vkuzo, and 1 more.
What were the notable Hugging Face updates?: v0.2.12, feat(vad): integrate Smart Turn v3.2 endpointing, and Merge pull request #429 from huggingface/agent/clean-dependency-extras.

KERNELS REFACTOR CLEANS UP LOADING PATH, TOKENIZERS KILLS DEPENDABOT NOISE

Hugging Face is untangling kernel loading internals across multiple repos while simultaneously eliminating 50% of its Dependabot burden through surgical dependency cleanup.

python 26 shipped

@huggingface Aug 4

LEROBOT BREAKS IMPORTS, FIXES TRAINING PIPELINE SILENTLY BREAKING

LeRobot 0.6.1 renames the types module and patches a critical bug where multi-stage training produced all-zero targets when episode annotations were malformed.

+11

python 69 shipped

@huggingface Aug 3

LEROBOT FIXES D405 REALSENSE STARTUP HANGS WITH AUTO-RECOVERY

LeRobot's camera stack now survives RealSense D405 hardware lockups that used to require a physical replug.

python 5 shipped

@huggingface Aug 2

LEROBOT CUTS EVAL OVERHEAD IN HALF, TRL FIXES DEVICE HARDCODING

LeRobot's evaluation loop was burning CPU cycles on finished episodes while waiting for slower sims to complete, and a fix shipping today reclaims that wasted compute.

python 18 shipped

Elsewhere on the wire

AI Agents Aug 6

LANGCHAIN HARDENS ANTHROPIC TOOL CALLING AND GIT CREDENTIALS AS OPENHANDS SHIPS 1.10.0

LangChain fixed a critical credential leak in evaluation runs while Anthropic tool schemas get proper fallback handling, as three major agent frameworks push releases across the ecosystem.

ai-agents 54 shipped

Local LLMs Aug 6

OLLAMA FIXES KV CACHE CORRUPTION, LLAMA.CPP HARDENS FILE SEARCH, VLLM KILLS MOE LEGACY CODE

Ollama patched a critical cache corruption bug that corrupts lazy snapshots during multi-request prefills, while llama.cpp hardened Windows junction traversal and vLLM removed 14 hardcoded MoE shards across quantization methods.