Supadchaya

Results 5 comments of Supadchaya

gpu version `int_nbit_split_embedding_codegen_lookup_function` not support `indices` and `offsets` with uint64 dtype

@tiankongdeguiji Hi, let me bring this to the FBGEMM team's attention and get back to you.

Can you add the Torch FBGEMM version to the requirements. txt file? Torch and fbgem are often incompatible

@imhandsome We'll work on it to highlight the requirement, but could you share the issues? Are you building from scratch or using FBGEMM package? Are you referring to nightly or...

fbgemm_gpu test fail

Hi @ywangwxd, can you show the result from running `nvidia-smi`?

Use Nova workflow to host all published wheel files at PyTorch site

Hi @atalman, sorry for pinging you again here. The build fails as it can't locate the file `setup.py`, which is in fbgemm_gpu subfolder. I did `cd` to the folder in...

ROCm symmem allreduce is slower than fbgemm car allreduce

cc: @q10 @ionuthristodorescu