PyTorch ships ExecuTorch WebGPU backend: run Llama 3.2 in-browser up to 7.4x faster
PyTorch · x · 2026-11-11
PyTorch's new ExecuTorch WebGPU backend lets you export once with torch.export and run the same .pte file in WebGPU browsers like Chrome and Safari — fully on-device, so prompts, photos and audio never leave the tab.
- Models verified: Llama 3.2, Qwen3, Gemma 3, SmolVLM, SAM 2, Depth Anything V2, D-FINE, Real-ESRGAN, Whisper, plus on-device LoRA/DiReFT fine-tuning.
- Benchmarks (Apple GPU, Chrome; vs CPU XNNPACK): Llama 3.2 1B prefill 5.4–7.4x faster (833 tokens/s at 512 context), decode 2.4–3.8x; Qwen3-0.6B decode up to 10x.
- Blog post and a WebGPU Playground are live for hands-on testing.
Related event: PyTorch Launches ExecuTorch WebGPU Backend for In-Browser Model Inference(2 posts)→
More from Infra
- Tokens are the wrong unit for the physical world, dev argues; vision compute will be priced in video-seconds and scenes — prateekj · 2026-11-11
- Report: Alice Huang leaned on her close ties with Jensen Huang to win Nvidia chip customers — GaryMarcus · 2026-11-11
- FAA's SMART system rerouted a 119-passenger flight around storms, saving 33 minutes — deanwball · 2026-11-11
- Qwen Image 2.1 runs 30-40% slower on Windows than Linux, user reports — wacomlover · 2026-11-11
- Commerce Dept delays entity list decision, Megaspeed can still buy advanced Nvidia AI chips — GarrisonLovely · 2026-11-11
- Liquid AI models are already running in Mercedes-Benz cars, phones next — agihouse_org · 2026-11-11