肥猫零零七

 It is recommended to add a new package to handle the 2.0 series separately, otherwise the existing processing for the 2.0 version will actually be run in the compatibility...

@YLGH en, use horovod don't need call dist.init_process_group(backend=backend),

@richardkiss @scw @dlumma @prabhuramachandran Whether there is a temporary solution to fix the problem

@walter-erquinigo @iamtimdavis I am sorry that I cannot upload the source code file, but I have located the problem, which is caused by special invisible characters in the """ """"...

@walter-erquinigo @iamtimdavis this is a crash demo [crash.mojo.txt](https://github.com/modularml/mojo/files/14716537/crash.mojo.txt)

> What is the meaning of these flags? > > Can they just be ignored for clangd's purposes? If so, you can instruct clangd to ignore it by creating a...

"arguments": ["/usr/local/cuda/bin/nvcc", "-c", "-ccbin=clang", "-Xcompiler", "-fPIE", "-O3", "-I/App/conda/envs/conda_xmake/include", "-I/usr/local/cuda/Common", "-I/usr/local/cuda/include", "-allow-unsupported-compiler", "-m64", "-rdc=true", "-gencode", "arch=compute_70,code=sm_70", "-gencode", "arch=compute_70,code=compute_70", "-o", "build/.objs/sgemm/linux/x86_64/release/src/cuc/02/sgemm/sgemm.cu.o", "src/cuc/02/sgemm/sgemm.cu"],

@HighCommander4 Is there a solution to this problem

@HighCommander4 I can't remove these flags because cuda c++ programs won't compile properly from the command line,i need it

@HighCommander4 The .clangd file should be stored in the directory of the engineering root directory in vscode project or below the .vscode directory