Jax
PulseAugur coverage of Jax — every cluster mentioning Jax across labs, papers, and developer communities, ranked by signal.
- used by Orbax Distributed Checkpointing With Jax 90%
- used by graphics processing unit 70%
- used by NVIDIA H100 70%
- used by Hugging Face Transformers 70%
- used by NumPy 70%
- used by robotics 70%
- instance of robotics 70%
- developed by Orbax Distributed Checkpointing With Jax 70%
- used by reinforcement learning 70%
16 day(s) with sentiment data
-
New JAX framework accelerates RL for penetration testing
Researchers have developed NASimJax, a new JAX-based framework designed to accelerate reinforcement learning (RL) for penetration testing. This framework significantly enhances the speed of existing simulators, enabling…
-
Weights & Biases streamlines ML experiment tracking with broad framework integration
Weights & Biases (W&B) offers a comprehensive platform for machine learning experiment tracking, logging metrics, configurations, and artifacts. The platform integrates with popular ML frameworks like PyTorch, TensorFlo…
-
New GPU-accelerated MPC solver TurboMPC achieves significant speedups
Researchers have developed TurboMPC, a novel model predictive control (MPC) solver designed for efficient execution on GPUs. This solver supports complex robotic applications by handling state and control inequality con…
-
New SOLAR framework automates deep-learning model performance analysis
Researchers have developed SOLAR, a new framework designed to automatically analyze the performance of deep-learning models. SOLAR calculates the theoretical minimum execution time for a given workload on specific hardw…
-
Superhuman AI agent dominates Generals.io using self-play RL
A new research paper details the creation of a superhuman AI agent for the real-time strategy game Generals.io. Trained for four days on high-end GPUs, the agent achieved the top rank among over 5,000 human players and …
-
Developer Builds Language Model From Scratch on MacBook
A developer detailed their experience building a language model from scratch using only a MacBook, eschewing GPUs and cloud services. This project provided insights into the inner workings of models like ChatGPT and hig…
-
New GARIP method enhances self-play convergence in zero-sum games
Researchers have introduced GARIP, a novel method for improving self-play in two-player zero-sum games. Unlike previous approaches that use fixed or periodically updated references, GARIP utilizes a running average of p…
-
Differentiable Atari 2600 emulator created for explainable AI research
Researchers have developed a fully differentiable emulator for the Atari 2600 Video Computer System, named jutari and jaxtari, implemented in Julia and JAX respectively. This system provides a complex, known ground trut…
-
Researcher seeks library to release new optimization algorithm
A researcher is seeking recommendations for the best library to release their newly developed QQN Quadratic Quasi-Newton optimization algorithm. They have existing implementations in Rust, Java, and JavaScript but want …
-
New CRAX benchmark accelerates safe reinforcement learning evaluations
Researchers have introduced CRAX, a new benchmark designed to accelerate the evaluation of safe reinforcement learning (RL) agents. Built using the MuJoCo XLA physics engine, CRAX offers up to a 100x speedup compared to…
-
New framework unifies neurosymbolic learning with neural networks · 2 sources tracked
Researchers have developed NeSyCat Torch, a novel framework that integrates categorical semantics with neural networks for neurosymbolic learning. This implementation, available in HaskTorch, JAX, and PyTorch, aims to u…
-
New Agentic Framework Automates PyTorch to JAX Deep Learning Model Migration
Researchers have developed an autonomous system to migrate deep learning models from PyTorch to JAX, a process typically manual and error-prone. Their framework combines In-Context Learning (ICL) with an oracle-driven s…
-
Google's DiffusionGemma LLM Achieves 1000 Tokens/Sec with Diffusion Architecture
Google DeepMind has released DiffusionGemma, an open-weight LLM that utilizes a diffusion architecture for text generation, enabling significantly faster inference speeds compared to traditional autoregressive models. T…
-
Agent-based models tuned to Lotka-Volterra dynamics
Researchers have developed a method to tune agent-based predator-prey models to better align with Lotka-Volterra dynamics. This approach uses a feature-based loss function to optimize environmental and demographic param…
-
AI models enhance wildfire prediction and suppression planning
Researchers have developed a new framework that combines deep learning with cellular automata to improve wildfire spread prediction and suppression planning. The first approach uses a hybrid CNN-cellular automaton model…
-
Compiler-first duality enables portable O(1) Mamba-2 inference
Researchers have developed a new method for optimizing Mamba-2 inference, focusing on compiler-first state space duality. This approach enables portable autoregressive caching with $O(1)$ complexity, eliminating the nee…
-
Hugging Face Transformers library simplifies AI model integration
The Hugging Face Transformers library has become a cornerstone for AI development, simplifying the process of loading and utilizing pre-trained models. Initially a chatbot startup, Hugging Face pivoted to open-source to…
-
GraphGP algorithm scales Gaussian processes to billion parameters
Researchers have developed GraphGP, a GPU-accelerated algorithm designed to make Gaussian processes more scalable. This new method utilizes Vecchia's approximation to reduce the computational complexity from cubic to li…
-
Paper catalogs 84 numeric formats for ML hardware consistency
A new paper introduces a comprehensive catalog of 84 numeric formats used in machine learning hardware, addressing the challenge of silent divergences when porting models across different accelerators. The catalog inclu…
-
New GPU solver AFSAT enhances pseudo-Boolean satisfiability
Researchers have developed Accelerated Fourier SAT (AFSAT), a new GPU-accelerated solver for pseudo-Boolean satisfiability problems. AFSAT builds upon a previous proof-of-concept, FastFourierSAT, by engineering a fully …