AI Briefing#jax
The latest AI and developer news about #jax, with the original source and a short summary.
The latest AI and developer news about #jax, with the original source and a short summary.
Visko's Orbis model achieves 1.07x playback speed on a single Cloud TPU slice using JAX and custom kernels.
An open-source repository has been released for serving the 320B MoE model GLM-5.3-Flash on Kaggle's free TPU v5e-8.
NVIDIA achieved 1,025 TFLOPs per GPU on GB200 for Dropless MoE training via Transformer Engine for JAX, improving throughput by approximately 10x.
ZeroModels, a collection of pretrained models built with Keras 3, has been released and is runnable on JAX, PyTorch, and TensorFlow.
Google has released Gemini-based TPU development agent tools that assist with PyTorch to JAX conversion and kernel optimization.
SpaceX has nearly completed development of its own C-based AI training stack V1.0, targeting a performance improvement of more than 10x over JAX.
NVIDIA and Google Cloud have expanded their joint developer community to over 100,000 members along with JAX and Dynamo resources.
Parax, a JAX-based parametric modeling library, has been updated to version v0.7.
Benchmark results were released for 17 local LLMs and 5 agent harnesses.
A summary of how running Claude Code as a multi-day autonomous agent can be used to implement scientific computing code.
Transformers models from Hugging Face can now be loaded and used directly in KerasHub.
xAI has released the weights and architecture of **Grok-1**, a MoE model with 314 billion parameters, under the Apache 2.0 license.
Hugging Face Diffusers now supports SDXL inference on Cloud TPU v5e via JAX, improving performance and cost efficiency.
Hugging Face's Diffusers library now supports Flax, enabling fast inference of Stable Diffusion on TPU environments.
This introduces a technical methodology for building an SOTA-level sentence embedding model using 1 billion training pairs and JAX/Flax.