SGLang publishes cookbook for running StepFun's Step-5-Preview model
Overview
SGLang (@sgl_project) on X shares a cookbook for running StepFun's Step-5-Preview model.
The guide describes a B300 setup using tensor parallelism of 4, FP8 quantization, TensorRT-LLM MoE, and MTP speculative decoding, with HiCache off and a bf16 KV cache. The post is the only report on this topic, and it does not include benchmark results.
Written by AI from the articles below · updated Oct 15, 9:05 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
SGLang@sgl_projectOfficialSGLang publishes a cookbook for StepFun's Step-5-Preview modelAISGLang (@sgl_project) shares a cookbook for running StepFun's Step-5-Preview model. The guide covers a B300 setup with tensor parallelism of 4, FP8 quantization, TensorRT-LLM MoE, and MTP speculative decoding, with HiCache off and bf16 KV cache.
Heat trend
Not enough continuous observations to show a trend yet.