[Release] bump v0.4.5.post1 with an absolute leading performance advantage

[zhyncs]

The SGLang team is pleased to announce the release of v0.4.5.post1. We compared the offline performance of vLLM 0.8.4 and SGLang 0.4.5.post1. The benchmark results show that SGLang outperforms vLLM in all cases, it leads by around 10% in most cases, with a maximum lead of 38% 🚀

ref
https://x.com/lmsysorg/status/1913064701313073656
#5514

[celsowm]

what was the secret sauce ?

[zhyncs]

We are Performance Hackers :)

[zhyncs]

I just recorded a video to demonstrate the process of SGLang's performance benchmark on H200, which shows slight fluctuations compared to the data released in yesterday's GitHub issue. Feel free to provide feedback at any time.😎

https://x.com/zhyncs42/status/1913351701324857849

[celsowm]

I gonna make a small benchmark video using my humble 3060 12gb but soon my workplace (rio de janeiro state attorney) will get 8xH100 and sglang gonna be our inference engine for sure !