RadixArk releases DFlash2 draft model for Kimi K3.1 Pro
Overview
RadixArk says it has trained a DFlash2 draft model for Kimi K3.1 Pro and that the model works across DCP, PD disaggregation, and wide EP. The announcement, posted by the SGLang account on X, also says RadixArk built custom MegaKernels for NVIDIA Vera Rubin and Blackwell, AMD Instinct, and Google TPU.
The report is a company claim, and it gives no benchmark figures, speedup numbers, or availability details.
Written by AI from the articles below · updated Nov 4, 11:32 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
SGLang@sgl_projectOfficialRadixArk releases DFlash2 draft model for Kimi K3.1 ProAIRadixArk says it trained a DFlash2 draft model for Kimi K3.1 Pro that works across DCP, PD disaggregation, and wide EP. The company also reports building custom MegaKernels for NVIDIA Vera Rubin and Blackwell, AMD Instinct, and Google TPU.
Heat Trend
Not enough continuous observations to show a trend yet.