Skip to main content

Share story

AI

ExecuTorch adds a browser GPU backend for PyTorch models

ExecuTorch adds a browser GPU backend for PyTorch models Image: Primary
PyTorch says its new ExecuTorch WebGPU backend runs PyTorch models locally on browser GPUs without converting them to another framework's format. Developers export a model to an ExecuTorch file that can run across WebGPU browsers, keeping inputs and outputs on the device. The project reports support for language, vision, speech and on-device fine-tuning workloads. In tests of quantized Llama 3.2 1B and Qwen3-0.6B in Chrome on an Apple M1 Pro, it reports 4-7× faster prompt processing and 2-10× faster token generation than ExecuTorch's WebAssembly CPU backend. The backend remains in early development, with further work planned on stability and model coverage.
Sources
In this story
Published by Tech & Business, a media brand covering technology and business. This story was sourced from PyTorch and reviewed by the T&B editorial agent team.
Back to Newswire
Keep reading
Full wire
Capital Infrastructure
Capital Infrastructure

Tendrils raises $30 million for parallel processor development

Tendrils says it has completed a $30 million financing round led by Benchmark to develop a general-purpose processor built around thousands of cores operating in parallel. Other participants include Visionaries Club and the UK Sovereign AI Fund.

Security Capital
Security Capital

Daylight raises $80M Series B led by Goldman Sachs Alternatives

New York-based Daylight raised an $80M Series B led by Goldman Sachs Alternatives, the Wall Street Journal reported. Sources put its valuation at approximately $400M. Daylight provides managed security through a platform that pairs AI agents with human experts.