SGLang-TorchTPU optimizes the deployment of Muse Spark Open.
The story so far
RadixArk Team released Muse Spark Open on TPU v7 with SGLang-TorchTPU, enabling PyTorch models to run in WebGPU browsers without converting framework formats, supporting multiple models such as Llama 3.2, Qwen3, and Gemma 3. The WebGPU backend allows PyTorch models to run in WebGPU browsers without converting framework formats. This backend is 4–7 times faster than ExecuTorch's CPU path (Webassembly + XNNPACK) on prefill and 2–10 times faster on decode. During deployment, developers serve .pte files on webpages via CDN. This backend is still in early development, and the team will make it faster, more stable, and expand operator, data type, and model coverage. All operations are performed locally, and data does not leave the device.
AI-written from the reports · Updated 8 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- PyTorch BlogOptimizing Muse Spark Open on TPU v7 with SGLang-TorchTPU
ExecuTorch WebGPU backend allows PyTorch models to run in WebGPU browsers without converting framework formats, supporting multiple models such as Llama 3.2, Qwen3, and Gemma 3.
Heat over time
- Now
- 3
- Peak
- 412 Nov, 06:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.