SGLang publishes one-GPU Qwen3.8-27B recipes, claims 206.1 tokens per second

Ying Sheng and Banghua Zhu's inference project added NVFP4 and DFlash2 recipes, with project-reported throughput of 206.1 tokens per second on one RTX 5090.

By · Published

SGLang adds NVFP4 and DFlash2 recipes for Alibaba's Qwen3.8-27B — RadixArk's open inference project targets Blackwell GPUs, with single-card configurations for the RTX 5090 and DGX Spark.

Ying Sheng and Banghua Zhu's inference project added NVFP4 and DFlash2 recipes, with project-reported throughput of 206.1 tokens per second on one RTX 5090.