Jax
PulseAugur coverage of Jax — every cluster mentioning Jax across labs, papers, and developer communities, ranked by signal.
13 day(s) with sentiment data
-
vLLM 0.33.0 mandates CUDA 13, adopts Model Runner V2, and boosts speed
vLLM has released version 0.33.0, introducing significant changes that require user attention for a smooth upgrade. The release mandates CUDA 13.0, discontinuing support for older versions and necessitating driver updat…
-
New XStack benchmark reveals AI agents struggle with ML repo migration
Researchers have developed XStack, a new benchmark designed to evaluate the ability of AI agents to migrate machine learning repositories between different frameworks and hardware ecosystems. The benchmark focuses on co…
-
LLMs demonstrate "Weight-Space Programming" by reading/writing neural network weights
A new research paper explores the capability of Large Language Models (LLMs) to directly read and write the weights of neural networks, a concept termed "Weight-Space Programming." Experiments with Claude Opus 5.5 sugge…
-
New evaluation protocol and embedding method for Physics-Informed Neural Networks
Researchers have introduced PINNeval, a new evaluation protocol and accompanying JAX package called PINNpipe, designed to systematically assess Physics-Informed Neural Networks (PINNs). This protocol addresses limitatio…
-
New OrbitalGym simulator explores impact of planning vs. system design in satellite games
This paper introduces OrbitalGym, a JAX-based simulation environment designed to study adversarial encounters between satellites. The environment allows for large-scale parallel simulations on GPUs, enabling researchers…
-
New KPPax tool enables differentiable machine learning for atmospheric chemistry
Researchers have developed KPPax, a new preprocessor that enables the use of atmospheric chemistry mechanisms written in the Kinetic PreProcessor (KPP) language within machine learning frameworks. KPPax generates code t…
-
Mirendil launches Fern 0.1 with 10M-token context window
Mirendil has released Fern 0.1, its inaugural AI model, featuring an extensive 10 million token context window. This model is designed for AI research and development tasks, particularly for processing large codebases a…
-
New paper reveals numerical issues in ML libraries, proposes fixes
A recent paper by Barron et. al. identified numerical issues in the softmax and cross-entropy functions used across common machine learning libraries like PyTorch and JAX. Researchers demonstrated that property-based te…
-
Deepity library offers faster predictive coding than JAX and ngc-learn
A new predictive coding library called Deepity has been released, offering a C++/CUDA implementation with Python bindings. This library aims to provide an alternative to traditional backpropagation by using local belief…
-
New JAXbounce benchmark tests active inference in complex environments
Researchers have introduced JAXbounce, a new benchmark designed to test active inference agents in continuous control environments that extend beyond simple, low-dimensional settings. This benchmark features continuous …
-
New AI research spans autonomous driving, economics, image denoising, and optimization
Researchers have developed DriveMind, a new framework for autonomous driving that uses multimodal large language models to track multiple scene hypotheses simultaneously, improving performance on ambiguous scenarios. Se…
-
Anyscale and Google Cloud enable native PyTorch on TPUs via Ray
Anyscale and Google Cloud have partnered to introduce native support for PyTorch on Google TPUs, leveraging the Ray distributed computing framework. This integration allows ML teams to manage distributed PyTorch trainin…
-
Hugging Face integrates native PyTorch support for Google TPUs
Hugging Face has integrated native support for Google's Tensor Processing Units (TPUs) through a new PyTorch backend called TorchTPU. This integration simplifies the process for developers, allowing them to run PyTorch …
-
Pearl-Chat: New Luganda Language Model Engineered from First Principles
Researchers have developed Pearl-Chat, a novel language model specifically engineered for the Luganda language, an under-represented African language. This 8.5 million-parameter model was built from the ground up using …
-
JAX gets new quasi-Newton optimization methods
A new implementation of the Self-Scaled Broyden family of quasi-Newton methods has been released for JAX, a numerical computation library. This implementation, built on the Optimistix library, includes various BFGS, DFP…
-
New Rondo testing method finds 117 bugs in deep learning compilers
Researchers have developed Rondo, a novel stateful testing method designed to uncover bugs in deep learning compilers. By representing test cases as model-invocation sequences, Rondo effectively identifies issues arisin…
-
New JAX library enables advanced hyperbolic deep learning
A new open-source library called hyperbolix has been released, designed for hyperbolic deep learning within the JAX framework and built upon Flax NNX. This library offers a unified interface for six different manifolds,…
-
VIDRAFT trains LLM from scratch on NPUs, bypassing GPUs
Korean AI startup VIDRAFT has successfully pre-trained a large language model using Neural Processing Units (NPUs) instead of traditional GPUs. This achievement challenges the prevailing industry assumption that GPUs ar…
-
Schedule-Free AdamW leads AlgoPerf V1 benchmark for deep learning training algorithms
A new benchmark called AlgoPerf V1 has evaluated deep learning training algorithms, focusing on wall-clock time-to-target across various workloads. Schedule-Free AdamW emerged as the top performer among 13 submissions i…
-
New coordinate-free AD operator integrates geometry into machine learning
Researchers have developed a new, coordinate-free automatic differentiation (AD) operator that can represent various geometric differential operations. This operator is designed to work seamlessly with existing AD frame…