WebGPU
PulseAugur coverage of WebGPU — every cluster mentioning WebGPU across labs, papers, and developer communities, ranked by signal.
6 day(s) with sentiment data
-
llama.cpp, PyTorch, and new MoE model see significant updates
The llama.cpp project has released updates enhancing WebGPU acceleration and simplifying FlashAttention implementation for more efficient local LLM inference. Concurrently, PyTorch's MPSInductor now supports unsigned in…
-
parakeet.wgsl brings fast, browser-based ASR via WebGPU
A new project called parakeet.wgsl enables fast and accurate automatic speech recognition (ASR) directly within a web browser. This implementation leverages raw WebGPU compute shaders and SIMD WebAssembly for audio proc…
-
Self-healing web scrapers use LLMs and Playwright to overcome site changes
Traditional web scrapers often fail due to minor website changes, but a new approach uses Large Language Models (LLMs) and Playwright to create self-healing agents. These agents can adapt to DOM mutations by visually an…
-
llama.cpp PR caches MoE experts for faster local AI inference · 4 sources tracked
A new pull request for llama.cpp introduces a method to cache frequently used Mixture of Experts (MoE) layers on the GPU, significantly boosting inference speeds for models like Qwen3.6-35B-A3B by up to 2x on consumer h…
-
Manim animation engine now runs in browsers via WebGPU
Manim, the animation engine developed by 3Blue1Brown, has been made accessible in web browsers through the use of WebGPU. This advancement allows for the rendering of complex mathematical animations directly within a we…
-
PrismML releases Bonsai 27B, a 27B LLM for offline mobile use
PrismML has released Bonsai 27B, a 27-billion-parameter large language model capable of running offline on mobile devices like the iPhone 17 Pro Max. The model achieves its small footprint through a novel 1-bit training…
-
Google launches LiteRT.js for in-browser AI model inference
Google has introduced LiteRT.js, a JavaScript binding for its on-device inference library, formerly known as TensorFlow Lite. This new tool allows .tflite models to run directly within web browsers, leveraging WebGPU, W…
-
Bonsai 27B: 1-bit LLM runs in browser, shrinks to 3.8GB
The PrismML team has released Bonsai 27B, a 1-bit quantized large language model that can run locally in a web browser. This quantization technique reduces the model's size from 54GB to 3.8GB, while reportedly maintaini…
-
Serverless AI architecture runs LLMs entirely in browser tab
A technical paper outlines a novel serverless AI architecture that runs entirely within a browser tab, eliminating the need for backend infrastructure. This approach leverages Java compiled to WebAssembly for business l…
-
Browser-based image inpainting model runs locally via WebGPU and ONNX
A 200 million parameter image inpainting model has been developed to run entirely within a web browser. This model utilizes WebGPU and ONNX technologies, requiring only 1.24GB of locally cached weights. The innovation e…
-
Kuma project compiles PyTorch models for browser execution via WebGPU
A new project called Kuma aims to compile PyTorch models into self-contained WebGPU executables. This approach would allow models to run directly in the browser without needing Python or a server-side runtime. The proje…
-
230M LFM2.5 model runs in-browser at 1,400 tokens/sec
A 230 million parameter model, LFM2.5, is now capable of running in a web browser at a speed of 1,400 tokens per second. This performance is achieved through custom WebGPU kernels, developed by individuals previously as…
-
SDXL image generation runs locally in browser via WebGPU
An open-source browser extension has been developed to run the SDXL image generation model locally using WebGPU. This extension allows users to install and run the model directly within their browser without complex set…
-
Moebius image inpainting model ported to browser using Claude Code
Simon Willison successfully ported the Moebius 0.2B image inpainting model to run in a web browser using Claude Code. The process involved converting the model to ONNX format and leveraging WebGPU for browser-based exec…
-
New AI frameworks enable safer multi-agent systems and browser-based latent collaboration
Researchers have developed a new framework for hierarchical multi-agent reinforcement learning that enforces safety constraints while maintaining efficiency. This approach uses a constraint manifold at a low level to en…
-
Gemma 4-E2B runs in-browser at 255 tok/s with WebGPU kernels
A demo and WebGPU kernels for Gemma 4-E2B have been released, enabling in-browser operation at approximately 255 tokens per second. The optimization was reportedly aided by Fable 5 before its shutdown. The release inclu…
-
Browser-based AI controls virtual hand in physics sandbox
A new AI sandbox called Semantic Hand allows users to control a virtual hand in a browser environment using natural language prompts. The system leverages local AI models like Nemotron 3 Nano 4B or Qwen 3.5-4B, running …
-
Chrome extension enables local Stable Diffusion 1.5 image generation
A new Chrome extension allows users to run Stable Diffusion 1.5 locally within their browser using WebGPU. This tool requires Chrome 113+ and approximately 4-6 GB of RAM, with an initial download of about 2.1 GB for the…
-
Browser-based real-time voice changer released as MVP
A developer has created a real-time voice changer that operates entirely within a web browser. This tool leverages WebAssembly, ONNX Runtime, and WebGPU for its functionality. The creator has released it as a minimum vi…
-
Chrome proposes on-device Embedding API for web apps
Google Chrome is proposing a new Embedding API that would allow web developers to generate vector embeddings for content directly on the user's device. This API aims to leverage on-device AI infrastructure to enable fea…