Together AI introduces the inference optimization roadmap for Together Compiler, noting that CUDA Graphs and fine-grained PDL can recover approximately 98% of the performance of hand-written megakernels.
Source:Together AI · Read original
Together AI introduces the inference optimization roadmap for Together Compiler, noting that CUDA Graphs and fine-grained PDL can recover approximately 98% of the performance of hand-written megakernels.
Source:Together AI · Read original
Leave a Reply