Tagged articles
JAX
21 articles · Page 1 of 1
Google researchers introduce T2J-Bench, a benchmark with 63 repository-level tasks evaluating coding agents on PyTorch-to-JAX migration, and TwinHarness, a harness that uses executable source code as feedback to improve migration fidelity, achieving an 11.4% relative score improvement over baseline agents.
JAXPyTorchT2J-Bench
0 likes · 14 min read

This tutorial walks through building and training a physics‑informed neural network (Euler‑PINN) in JAX to simulate transonic airfoil flow, detailing the governing Euler equations, structured‑grid coordinate mapping, loss formulation, L‑BFGS optimization, and a thorough comparison with a high‑resolution finite‑volume reference solution that highlights both accuracy and efficiency trade‑offs.
Euler-PINNJAXL-BFGS
0 likes · 20 min read

This tutorial demonstrates how to solve the two‑dimensional steady incompressible Falkner‑Skan viscous boundary‑layer equations using a physics‑informed neural network implemented in JAX, detailing the network architecture, loss construction, two‑stage training strategy, and achieving sub‑0.01 % relative errors with only 1 % interior collocation points.
Falkner‑SkanJAXPINN
0 likes · 17 min read

This tutorial presents a complete DeepONet workflow—two‑step separated training, Rowdy activation, SVD orthogonalisation, and a 10‑member ensemble—that predicts the density, velocity and pressure fields of the one‑dimensional Sod shock‑tube problem with an average test‑set relative error of 2.23% after only 22 minutes of training on an RTX 4090.
DeepONetJAXNeural Operator
0 likes · 22 min read

This tutorial presents a comprehensive JAX implementation and analysis of four neural‑operator methods—standard DeepONet, POD‑DeepONet, 2‑D Fourier Neural Operator, and time‑stepping 1‑D FNO—applied to the 1‑D advection equation, comparing their mathematical foundations, parameter counts, training efficiency, and prediction accuracy.
DeepONetFNOJAX
0 likes · 20 min read

This tutorial demonstrates how continuous physics‑informed neural networks (PINNs) combined with stream‑function parameterization and nested forward‑mode automatic differentiation (JVP) can accurately identify the convection and viscosity coefficients of a two‑dimensional Navier‑Stokes cylinder‑wake problem from sparse velocity observations, achieving sub‑0.2% error for the convection term and robust performance even with 1% measurement noise, all within a few minutes on a single RTX 4090 GPU.
JAXNavier-StokesPINNs
0 likes · 28 min read

This tutorial presents the Time‑Induced Neural Network (TINN) framework that overcomes the time‑entanglement issue of standard PINNs by introducing a dedicated time‑subnet with FiLM modulation, employs a Levenberg‑Marquardt optimizer for second‑order updates, and demonstrates a 1e‑6 relative error solution of the 1‑D viscous Burgers equation in just 371 seconds on an RTX 4090.
Burgers EquationFiLM ModulationJAX
0 likes · 21 min read

The tutorial introduces Scale‑PINN, which adds an evolutionary regularization term inspired by pseudo‑time stepping to the PINN loss, enabling a shared‑backbone network to solve the lid‑driven cavity Navier‑Stokes problem at Re = 7500 in about 100 seconds and reducing the relative velocity error by roughly 96 % compared with a standard PINN.
Evolutionary regularizationHigh Reynolds numberJAX
0 likes · 25 min read

Training Physics‑Informed Neural Networks (PINNs) often stalls due to gradient‑flow pathology, but a gradient‑adaptive weighting scheme dramatically rebalances loss contributions, as demonstrated on a nonlinear Klein‑Gordon equation where error drops by 95.5% compared with standard PINNs.
JAXKlein-GordonPINN
0 likes · 24 min read

This article explains why TPU programming differs from GPU, describes the explicit HBM‑VMEM‑register data movement required on TPU, introduces the Pallas grid‑BlockSpec‑Ref model, and walks through four progressively more complex kernels—including element‑wise add, tiled dot product, fused RMSNorm with scratch memory, and a production‑grade FlashAttention implementation—showing how each kernel maps to the TPU memory hierarchy and leverages Pallas features such as input_output_aliases and PrefetchScalarGridSpec.
FlashAttentionJAXPallas
0 likes · 20 min read

This tutorial explains how a simple coordinate scaling (VS‑PINN) reduces stiffness in physics‑informed neural networks, demonstrates its implementation in JAX for the 2D steady incompressible Navier‑Stokes cylinder‑flow benchmark, and shows that after 80 000 Adam iterations the relative errors drop to 2.10 % (u), 5.06 % (v) and 4.45 % (p).
JAXNavier-StokesPINN
0 likes · 24 min read

This tutorial explains why standard physics‑informed neural networks fail on high‑frequency partial differential equations due to spectral bias, and demonstrates how random Fourier feature embeddings, multi‑scale concatenation or spatio‑temporal separation, and Neural Tangent Kernel‑based adaptive loss weighting together overcome the bias and achieve accurate, stable solutions for heat, Poisson, and wave equations using JAX.
Fourier FeaturesJAXMulti-Scale
0 likes · 23 min read

This tutorial explains how the Neural Tangent Kernel (NTK) perspective reveals the loss‑balance problem in Physics‑Informed Neural Networks (PINNs), introduces an NTK‑based adaptive‑weight algorithm, provides a full JAX implementation for a 1‑D high‑frequency wave equation, and shows that input normalisation dramatically improves accuracy while only modestly increasing training time.
JAXNTKPINN
0 likes · 27 min read

The article examines Google’s AI resurgence after Sergey Brin’s comeback, detailing the evolution of TPU hardware from v1 to v7, the strategic focus on algorithmic efficiency, comparisons with Nvidia’s B200, the role of JAX/XLA, and how these advances create a powerful competitive moat for Google’s AI infrastructure.
AI hardwareGoogle TPUJAX
0 likes · 8 min read


AntTechNov 4, 2025 · Artificial Intelligence
SGLang-Jax is a cutting‑edge, fully Jax‑based open‑source inference engine that delivers native TPU performance, integrates advanced features like continuous batching, tensor and expert parallelism, and speculative decoding, while providing detailed installation and usage guidance for developers.
JAXSGLang-JaxTPU inference
0 likes · 10 min read

This article introduces Google’s Tunix library for JAX‑based LLM post‑training, explains its core features such as supervised fine‑tuning, reinforcement learning and knowledge distillation, and provides detailed installation steps and a complete TPU‑accelerated QLoRA fine‑tuning workflow on the Gemma 2B model, including code snippets and inference testing.
AIJAXLLM
0 likes · 8 min read


21CTOMar 18, 2024 · Artificial Intelligence
Elon Musk’s AI startup xAI has open‑sourced its 314‑billion‑parameter Grok‑1 model, detailing its Rust‑based, JAX‑powered architecture, extensive parameter count, training data limits, licensing terms, hardware requirements, and community reactions, offering developers unprecedented access to a competitive large‑language‑model framework.
AIGrok-1JAX
0 likes · 9 min read

Google has released the open‑source Gemma LLM series (2B and 7B parameters) built on Gemini‑style architecture, offering free, commercial‑ready models that run on notebooks, support JAX/PyTorch/TensorFlow, outperform many open‑source peers, and have quickly sparked extensive community testing and discussion.
GemmaGoogleJAX
0 likes · 5 min read

The article demonstrates how to evaluate, compress, and convert deep‑learning models for edge devices using TensorFlow, JAX, and TVM—showing a faster iPhone‑based MNIST training benchmark, FLOPs measurement scripts, TFLite/ONNX/CoreML conversion, TVM compilation with auto‑tuning, and up to 50 % speed improvements on mobile NPU hardware.
JAXTVMTensorFlow
0 likes · 29 min read

This article examines the computational capabilities of modern mobile devices for machine learning, compares training times on a MacBook and iPhone, explains model evaluation metrics like FLOPs, and provides step‑by‑step guides for converting and optimizing models using TensorFlow, PyTorch, ONNX, JAX, and TVM for edge deployment.
JAXTVMTensorFlow
0 likes · 29 min read

This article introduces Google’s JAX library, covering its origins, core features such as automatic differentiation, JIT compilation, parallel and vectorized mapping, installation steps, code examples, and a comparative overview with TensorFlow and PyTorch for deep‑learning practitioners.
GPUJAXPyTorch
0 likes · 11 min read
