ehion
ehion
项目owner其实把该提供的信息都提供了,剩下的就是你自己需要做的工作了。
@abdoelsayed2016 totally agree with u,this project does not mention the important thing,they just give simple code to show Chinese llama result but not the process they do。 As your problem...
@denera https://github.com/NVIDIA/TransformerEngine/issues/1583 hello, i'am curious about the implementation of atomic_gemm_overlap_ag , how does it make the consistence of AG output between devices in TP group? @yaox12