SGLang releases open-source Muse Spark Open recipe for TPU, with RadixArk and Google work
Overview
SGLang announces that the code for running Muse Spark Open on SGLang-TorchTPU is open source, and links a cookbook recipe on its documentation site for setting it up.
In a separate post, SGLang says RadixArk is working with Google on a fully pipelined TPUv7 megakernel for Muse Spark Open, optimized for long-context agentic workloads. SGLang says TPUv7 performance is now very competitive with popular GPU options, and that more models on SGLang-TorchTPU are planned. The post does not include benchmark figures.
Written by AI from the articles below · updated Nov 5, 6:10 PM ET
Check the sources:
Developments
2 developments
- Nov 5, 6:08 PM ET · 1 articleSGLang shares a Muse Spark Open recipe for SGLang-TorchTPUSGLang: SGLang shares open-source Muse Spark Open recipe for SGLang-TorchTPU
- Nov 5, 6:08 PM ET · 1 articleRadixArk and Google build a TPUv7 megakernel for Muse Spark OpenSGLang: RadixArk and Google build a TPUv7 megakernel for Muse Spark Open
Article timeline
The articles in this story. Times are ET.
SGLang@sgl_projectOfficialSGLang shares open-source Muse Spark Open recipe for SGLang-TorchTPUAISGLang says the code for running Muse Spark Open on SGLang-TorchTPU is open source. It links a cookbook recipe on its documentation site for setting it up.
SGLang@sgl_projectOfficialRadixArk and Google build a TPUv7 megakernel for Muse Spark OpenAISGLang says RadixArk is working with Google on a fully pipelined megakernel for Muse Spark Open, optimized for long-context agentic workloads. SGLang says performance on TPUv7 is now very competitive with popular GPU options. More models on SGLang-TorchTPU are planned.
Heat Trend
Not enough continuous observations to show a trend yet.