@ekzhang1: Some progress on jax-js this weekend! - New matmul benchmarks - Real-time TTS demo (http://jax-js.com/tts) is much fast…
Summary
Progress on jax-js includes new matmul benchmarks, faster real-time TTS demo, improved codegen, and an LLM demo running Gemma 3 270B. jax-js is an open-source web ML framework.
View Cached Full Text
Cached at: 05/18/26, 04:34 PM
Some progress on jax-js this weekend!
- New matmul benchmarks
- Real-time TTS demo (https://t.co/J9MuFkrxpm) is much faster, codegen improved to avoid GPU kernel recompiles
- Finally an LLM demo (https://t.co/iVACirOplm) with Gemma 3 270B - actually faster than I thought https://t.co/i6HAdcVTBD
jax-js – Open-source web ML framework
Source: https://jax-js.com/tts Kyutai Pocket TTS (Web)## Kyutai Pocket TTShttps://github.com/ekzhang/jax-js/tree/main/website/src/routes/tts
Text-to-speech AI voice model, running in your browser withjax-js.
Advanced optionsSeed
Temperature: 0.70
LSD Decode Steps: 1
Similar Articles
jax-js (Website)
jax-js is an open-source ML library and compiler for the web, bringing JAX-like numerical computing to JavaScript with WebGPU and WebAssembly kernels, all running client-side.
Mahjax: A GPU-Accelerated Mahjong Simulator for Reinforcement Learning in JAX
This paper introduces Mahjax, a fully vectorized Riichi Mahjong simulator implemented in JAX for GPU-accelerated reinforcement learning, achieving high throughput and enabling tabula rasa training.
@_DivyaMakkar: Spent some time creating a lightweight, performant async RL stack in pure JAX with @AdityaMakkar0! We share a work log …
The authors have created a lightweight, performant asynchronous reinforcement learning stack in pure JAX, sharing a work log with insights on inference, RDMA weight transfer, memory optimizations, and sharding for scaling RL systems.
@maxencefaldor: Interested in emergence, self-organization or morphogenesis? Introducing CAX: Cellular Automata Accelerated in JAX, a h…
CAX is a high-performance open-source library for cellular automata research, built on JAX. It accelerates simulations up to 2000x and supports discrete and continuous automata in arbitrary dimensions.
PowerZooJax: A JAX-based Power System Benchmark for Reinforcement Learning
Researchers from Nanyang Technological University, Cornell, and Bristol introduce PowerZooJax, an open-source JAX-based benchmark suite for reinforcement learning in power system operation, providing five constrained MDP tasks (generation, transmission, distribution, DERs, data center microgrids) that run fully on GPU with substantial speedups over CPU-based simulators.