A high-engagement LocalLLaMA post shared reproducible benchmark data showing Qwen3.5-122B NVFP4 decoding around 198 tok/s on a dual RTX PRO 6000 Blackwell system using SGLang b12x+NEXTN and a PCIe switch topology.
A LocalLLaMA thread pulled attention to DFlash, a block-diffusion draft model for speculative decoding whose paper claims lossless acceleration above 6x and direct support for vLLM, SGLang, and selected Transformers backends.
NVIDIA AI Developer says a collaboration with SGLang achieved up to 25x faster DeepSeek R1 inference on GB300 NVL72 versus H200 and an 8x GB200 NVL72 gain within months. The post attributes gains to NVFP4 precision, disaggregation, and communication-compute overlap.