DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles

30 points by mji6 hours ago

1 comments

Palmik1 hour ago
Similar article for vLLM: <a href="https://vllm-website-pdzeaspbm-inferact-inc.vercel.app/blog/deepseek-v4" rel="nofollow">https://vllm-website-pdzeaspbm-inferact-inc.vercel.app/blog/...</a><p>Bechmarks from InferenceX (they do not have apples-to-apples setups to compare the different engines for whatever reason): <a href="https://inferencex.semianalysis.com/inference?i_hc=1&g_model=DeepSeek-V4-Pro&g_rundate=2026-04-25&g_runid=24943464864&i_prec=fp4%2Cfp8" rel="nofollow">https://inferencex.semianalysis.com/inference?i_hc=1&g_model...</a><p>I find it odd that sglang, vLLM, TRTLLM don't seem to want to publish benchmarks comparing each other. They used to, but now there seems to be some unspoken rule against it.<p>At least we get comparison against "other OSS engine" this time, but that could be HF's Transformers as well :)
- imjonse39 minutes ago
  They're OSS projects in a friendly competition, both working towards the goal of having alternatives to big closed source players. No need for jabs.
  - Palmik13 minutes ago
    I don't think "friendly" and "publishing benchmarks" are at odds with each other.<p>Model makers (both open and closed weight) typically publish benchmarks against other models and when they do not, people rightfully call them out.<p>Including comparison against "other OSS engine" is just not helpful (what if it's a sandbagged baseline like HF Transformers?)