# ByteDance explains faster open-model inference on Hopper with SGLang

> Original title: 5 | ByteDance

- Source: LMSYS Org
- Published: 2026-11-04T22:36:23.000Z
- Bad Influence: https://badinfluence.news/items/sz154skovgkgrsuuu1e9br0ab
- Original: https://x.com/lmsysorg/status/2118110135822516615

## Summary

ByteDance engineer Ho-Ren (Jack) Chuang shares the attention, MoE, and speculative decoding work behind faster open-model inference on Hopper GPUs using SGLang. The post covers these three areas, which the source does not quantify with speedup figures.

## Post

5 | ByteDance
Making Frontier Open Models Fast on Hopper with SGLang

Ho-Ren (Jack) Chuang shares the attention, MoE, and speculative decoding work behind faster open-model inference on Hopper.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHWUB_9xawAAQsT9.jpg&mode=thumb&exp=1794096000&sig=657f26d8c445bf55)