Zhijian Liu (@inco_ai) on DFlash and the other optimizations behind Inco’s open-model endpoints.
A look at its block-diffusion drafter, now in SGLang, and where the rest of the speed comes from.
Zhijian Liu explains DFlash and other speedups behind Inco's open-model endpoints
AIZhijian Liu of Inco describes DFlash, a block-diffusion drafter now available in SGLang, and other optimizations behind Inco's open-model endpoints. The post also covers where the rest of the speed comes from.
Source details
LMSYS Org · x.comPublished
Original title: Zhijian Liu (@inco_ai) on DFlash and the other optimizations behind Inco’s open-model endpoints.