All the benchmarks have either VLLM or sglang but not both. I assume this is to avoid twitter fights. I think a lot of people at labs like me are on a fork of one and spend some energy trying to translate one to the other, often poorly. I think it would be great for the community if you published both numbers (or at least encouraged it). Clearly this effort has sparked AMD to do a lot more useful shit so I think it VLLM v sglang competition would also probably be good for the community. I would also not allow disable_prefix_cache=True type things but I feel less strongly about that. I also think from VLLM and SGLANG people twitter wars, the fights are not that bad (more publicity) and could actually be resolved more gracefully on this platform.
中文说明
建议为每个模型/硬件配置同时发布 vLLM 和 SGLang 两套基准测试结果，而非仅选择其中一个框架。这有助于社区用户（尤其是使用其中一个框架分支的实验室）更好地对比两者性能，也能促进框架间的良性竞争。同时建议禁止 disable_prefix_cache=True 等不公平配置。
All the benchmarks have either VLLM or sglang but not both. I assume this is to avoid twitter fights. I think a lot of people at labs like me are on a fork of one and spend some energy trying to translate one to the other, often poorly. I think it would be great for the community if you published both numbers (or at least encouraged it). Clearly this effort has sparked AMD to do a lot more useful shit so I think it VLLM v sglang competition would also probably be good for the community. I would also not allow disable_prefix_cache=True type things but I feel less strongly about that. I also think from VLLM and SGLANG people twitter wars, the fights are not that bad (more publicity) and could actually be resolved more gracefully on this platform.
中文说明
建议为每个模型/硬件配置同时发布 vLLM 和 SGLang 两套基准测试结果，而非仅选择其中一个框架。这有助于社区用户（尤其是使用其中一个框架分支的实验室）更好地对比两者性能，也能促进框架间的良性竞争。同时建议禁止
disable_prefix_cache=True等不公平配置。