Who contributed to Hugging Face on August 1, 2026?

14 developers shipped this update, including rian-dolphin, alejandro-ao, albertvillanova, AmineDiro, qgallouedec, Recoordinate, Ashmitha R, and BenjaminBossan, and 6 more.

What were the notable Hugging Face updates?

Cache Anthropic prompts instead of re-billing every turn, Fix Anthropic subscription login and OAuth token refresh, and Cache Anthropic prompts instead of re-billing every turn (#502).

@huggingface

Transformers, Datasets, and the open AI-model layer

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

TAU FIXES ANTHROPIC BILLING LEAK, UNLOCKS PROMPT CACHING

By RepoJournal · Filed 06:59 UTC on August 1, 2026 · About Hugging Face

14 people shipped this

rian-dolphin @rian-dolphin 3 cited

imstevenpmwork @imstevenpmwork 2 cited

albertvillanova @albertvillanova 1 cited

qgallouedec @qgallouedec 1 cited

ovuruska @ovuruska 1 cited

moon-bot-app @moon-bot-app 1 cited

alejandro-ao @alejandro-ao

AmineDiro @AmineDiro

Recoordinate @latent-9

Ashmitha R @Ashmitharaja

BenjaminBossan @BenjaminBossan

dependabot @dependabot

github-actions[bot] @github-actions[bot]

moon-bot-app[bot] @moon-bot-app[bot]

Tau shipped critical fixes for Anthropic subscriptions today, plugging a prompt caching hole that was re-billing identical system prompts on every single turn.

Tau 0.3.5 landed with the caching layer developers have been waiting for [1]. The library had wired up the *reporting* side of Anthropic's prompt cache (parsing cache_read, cache_write, and ephemeral_1h_input_tokens from responses) but never actually *requested* caching in payloads, so every turn re-billed the system prompt, tool schemas, and conversation history as fresh input [3]. In parallel, rian-dolphin patched three separate OAuth bugs blocking Anthropic subscription login entirely: a typo in the client ID where one character stood between users and "Invalid client id" errors, broken token refresh that would kill credentials overnight, and auth URL handling that failed under load [2]. On the TRL front, GRPO, DPO, KTO and RLOO now respect frozen reference adapters when working with PEFT's `target_parameters` on peft>=0.20.0, instead of always falling back to the base model [4]. Separately, DPO now rejects unsupported f_divergence_type combinations that were either silently ignored or half-wired at init time [5]. LeRobot merged robotics-specific image augmentations (8 transforms for real-world robustness), fixed a double-reset bug in LiberoEnv that paid two full resets per episode termination, and improved Hub checkpoint diagnostics for legacy models [6][7]. The huggingface.js inference client shipped text-to-speech support for DeepInfra and reordered conversational payloads to put `model` first, enabling prefix-based routing for proxies instead of forcing buffering to identify providers [8][9].

Action items

→ Upgrade to Tau 0.3.5 if you're running Anthropic subscriptions to enable prompt caching and fix OAuth token refresh huggingface/tau [immediate]
→ Update PEFT to 0.20.1 if training with target_parameters adapters to support frozen reference adapters in GRPO/DPO huggingface/peft [plan]
→ Pull LeRobot's latest if you're using LiberoEnv for real-world robotics evaluation; fixes double-reset overhead huggingface/lerobot [plan]

References

[1] Cache Anthropic prompts instead of re-billing every turn ↗ huggingface/tau
[2] Fix Anthropic subscription login and OAuth token refresh ↗ huggingface/tau
[3] Cache Anthropic prompts instead of re-billing every turn (#502) huggingface/tau
[4] Use a reference adapter with `target_parameters` on `peft>=0.20.0` ↗ huggingface/trl
[5] Reject unsupported `f_divergence_type` / loss type combinations in DPO ↗ huggingface/trl
[6] feat(transforms): add 8 robotics-relevant image augmentations ↗ huggingface/lerobot
[7] fix(libero): don't reset inside step() on termination (#4273) huggingface/lerobot
[8] [Inference Providers] deepinfra: add text-to-speech support ↗ huggingface/huggingface.js
[9] Serialize `model` first in conversational payloads ↗ huggingface/huggingface.js

Quick answers

What shipped in Hugging Face on August 1, 2026?: Tau shipped critical fixes for Anthropic subscriptions today, plugging a prompt caching hole that was re-billing identical system prompts on every single turn. In total, 45 commits, 44 pull requests, and 2 releases landed.
Who contributed to Hugging Face on August 1, 2026?: 14 developers shipped this update, including rian-dolphin, alejandro-ao, albertvillanova, AmineDiro, qgallouedec, Recoordinate, Ashmitha R, and BenjaminBossan, and 6 more.
What were the notable Hugging Face updates?: Cache Anthropic prompts instead of re-billing every turn, Fix Anthropic subscription login and OAuth token refresh, and Cache Anthropic prompts instead of re-billing every turn (#502).

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

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.

python 57 shipped

@huggingface Aug 5

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

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.