ppanchad-amd

Results 674 comments of ppanchad-amd

Megatron training adaptation issue, obtaining CUDA_SOME as None

Hi @JoJo-Lorray. Internal ticket has been created to investigate your issue. Thanks!

[Issue]: Does not honor cmake BUILD_SHARED

Hi @trixirt. Internal ticket has been created to investigate this issue. Thanks!

[Issue]: Does not honor cmake BUILD_SHARED

Closing ticket as PR merged into develop branch.

[Issue]: hardcoded toolchain paths break on Fedora and OpenSUSE

Hi @trixirt. Internal ticket has been created to fix this issue. Thanks!

[FAQ] what is CUDA GPU direct nvme equivalent thing in ROCm

@gaowayne Apologies for the delayed response. 1: Currently ROCm does not have a GPUDirect equivalent. 2: RDMA is supported by ROCm, but there is no NVMEOF-specific code in ROCm. They’d...

[FAQ] what is CUDA GPU direct nvme equivalent thing in ROCm

@gaowayne Can we close this ticket? Thanks!

[FAQ] what is CUDA GPU direct nvme equivalent thing in ROCm

@gaowayne Closing ticket since there is no response for a while. If you still require assistance, please re-open the ticket. Thanks!

[Question] Register data layout for two consecutive GEMMs in flash attention kernel (how is TransposedC implemented)?

@bulffi Internal ticket has been created to assist with your question. Thanks!

Fp64 xdlops gemm, some config give error result on rocm 5.1

@ltqin Please advise if we can go ahead and close this ticket as fixed. Thanks!

[Issue]: amd-smi reports different "size" value for XCD memory size different from the docs on MI300A at TPX and CPX

Hi @amroakmal. Internal ticket has been created to assist with your issue. Thanks!