Microsoft GraphRAG official releases
v3.3.0 fixes SQLite performance issue
Version 3.3.0 includes a fix for an SQLite performance issue.
The complete AI news desk
The AI world, in one verified news room.
Models, agents, robotics, research, business and policy, with the original source attached. Search and move beyond the latest thirty updates.
Open the 30-day AI Change IndexLatest and archive
Open the fast flashes streamModels
whisper.cpp official releases
Whisper version bumped to 1.9.5 in maintenance release.
The update includes a bump to ggml v0.26.0 and fixes for CUDA and Vulkan.
Hugging Face TRL official releases
Patch 1.14.2 corrects generation path errors in GRPO, RLOO, and Distillation.
The patch release v1.14.2 corrects generation path issues in GRPO, RLOO and Distillation. It also fixes metrics errors with vLLM.
Hugging Face PEFT official releases
PEFT v0.21.2 fixes encoder-decoder model issue in Transformers ≥ 5.18.0
The PEFT release v0.21.2 resolves a bug affecting encoder-decoder models with Transformers 5.18.0.
Hugging Face TRL official releases
vLLM server mode now generates one completion per sample after tool calls.
The v1.14.1 release fixes a server-mode crash during initial weight sync on vLLM 0.20 to 0.25.
Hugging Face PEFT official releases
Tensor Parallel Now Compatible with PEFT in v0.21.1
The v0.21.1 release supports Tensor Parallel with PEFT. It requires Transformers version 5.17.0 or higher.
Hugging Face TRL official releases
TRL v1.14.0 drops six experimental trainers, adds Triton kernel for log-probs and entropy.
The v1.14.0 release of TRL removes six experimental trainers, including GRPOWithReplayBufferTrainer and BCOTrainer.
Flax official releases
Flax 0.12.10 released with Flax RoPE verification change
Flax 0.12.10 now uses attribute lookup for HiPrim fallback.
Microsoft GraphRAG official releases
Derek Worthen changed default package feed index in v3.2.0
v3.2.0 includes a new cache implementation by gaudyb.
Mistral Common official releases
New version 1.12.0 adds agnostic validation mode.
The release includes a new agnostic validation mode and updated documentation tooling to version 0.12.6.
ModelScope official releases
v1.40.1 adds two new speech separation models for 8 kHz audio
v1.40.1 includes two new speech separation models for 8 kHz audio. It also adds SDK and CLI compatibility for agent identity management.
PaddlePaddle official releases
PaddlePaddle 3.4.0 adds a high-performance Muon optimizer for large-scale distributed training.
PaddlePaddle 3.4.0 lists a Muon optimizer for distributed training and a uniform-generation alignment fix for large tensors.
whisper.cpp official releases
The ggml-metal.metal file has been removed from the metal repository.
This commit removes the obsolete ggml-metal.metal file, clearing a path for the whisper.cpp release.
whisper.cpp official releases
Microsoft releases experimental Windows Arm64 CUDA build
The Windows arm64 (CUDA) build of whisper-bin-win-cuda is experimental, using a preview edition of the CUDA Toolkit.
PyTorch Lightning official releases
PyTorch Lightning 2.6.6 restricts _instantiator hyperparameter to prevent code execution.
Lightning 2.6.6 restricts _instantiator hyperparameter to prevent code execution.
OpenCV official releases
OpenCV 5.0.0 includes Android SDK with 16kb-page-fix for Google Play releases.
OpenCV 5.0.0 now supports Windows-on-Arm binaries validated on Qualcomm Snapdragon devices.
Swiss Data Science Center Renku official releases
Renku 2.20.1 fixes session log collection for anonymous sessions.
Renku 2.20.1 resolves a backend issue affecting session log collection during anonymous sessions.
Mideind GreynirEngine official releases
Icelandic parser adds support for quoted questions and deterministically selects parse trees.
The update includes new parsing for Icelandic verb structures and quoted questions.
Swiss Data Science Center Renku official releases
Renku 2.20.0 adds project storage shared among all users of a project.
Renku 2.20.0 adds project storage shared among users, with administrators setting size limits.
OpenCompass official releases
OpenCompass 0.5.4 adds LiteLLM as an AI gateway backend.
OpenCompass v0.5.4 adds LiteLLM as an AI gateway backend and includes new benchmarks for finance and code reasoning.
tinygrad official releases
tinygrad 0.14.0 adds Qwen kernels, tool calling and parallel compilation
The release also speeds tokenization and USB copy-in, and removes Tensor.training and Tensor.train().
PyTorch official releases
CPU Reverts to 2.13 in Latest Release Candidate
Surprise rollback in latest release—what's changed under the hood?
PyTorch official releases
PyTorch Prepares Header-Only Atomic Addition in Latest Release Candidate
PyTorch's latest update hints at a hidden efficiency boost—find out what's changed under the hood.
You reached the end of the available verified archive.