Wei Wu

module envs are missing when using SlurmSpawner

2

### Bug description My sbatch script is: ``` [I 2022-06-02 22:44:23.718 JupyterHub batchspawner:292] Spawner submitted script: #!/bin/bash #SBATCH -t 00:30:00 #SBATCH -N 1 env module load PrgEnv/GCC+OpenMPI/2021-05-27 cuda gcc openmpi...

Compilation issue on the PowerPC

8

I try to use TBLIS on the PowerPC, and I am able to compile the TBLIS, but when I try to include a header file, I am getting the following...

BERT in Megatron-LM-v1.1.5-3D_parallelism does not support pipeline parallelism

1

I try to run the BERT with pipeline parallelism, but I get an error: File "DeepSpeedExamples/Megatron-LM-v1.1.5-3D_parallelism/pretrain_bert.py", line 146, in args_defaults={'tokenizer_type': 'BertWordPieceLowerCase'}) File "/DeepSpeedExamples/Megatron-LM-v1.1.5-3D_parallelism/megatron/training.py", line 81, in pretrain model, optimizer, lr_scheduler...

hipDNN fails to compile in nvcc platform

1

I tried to compile hipdnn on Nvidia platform, and I get the following errors: ``` /home/wwu/app/hipDNN-NV/library/src/nvcc_detail/hipdnn_cudnn.cpp:2349:66: error: macro "CHECK_CUDNN" passed 7 arguments, but takes just 1 &cuTF, &k, &c, &h,...

SaveAsPriorBuffer is not working if dData is offset of hipMalloced buffer

2

In https://github.com/ROCmSoftwarePlatform/hipDNN/blob/master/library/src/hcc_detail/hipdnn_miopen.cpp#L105 ``` CHECK_HIP(hipMemPtrGetInfo( dData, &dPriorSize)); // Get the info of the gradient dx size ``` If the dData is not directly allocated by hipMalloc, instead, it is part of...