int21_ai 称 Rust 推理栈在 DeepSeek V4.1 Flash 上超 1700 tok/s

Bing Xu @bingxu_原推发布于 2026.10.23

在 X 上打开原推

This shows just how challenging it was to build @int21_ai 's Rust Inference Factory. We generated the entire stack and outperformed vLLM and SGLang across 20+ models in ultra-fast mode, including achieving over 1,700 tokens/s on DeepSeek v4.1 Flash inference.