link89

#!/bin/bash
set -e

[ -f build.done ] || {

module purge
module add compiler/2022.0.1
./configure CC=icc --prefix=/data/share/lib/libbeef/latest --enable-optmax
make
sudo make install

link89

#/bin/bash
set -e

# Hello world test
cat <<EOF > hello.cu
#include <stdio.h>

__global__ void helloFromGPU(void) {
    printf("Hello World from GPU!\n");
}

link89

#!/bin/bash
set -e

# reset environment
conda deactivate || true
module purge

export OPENBLAS_NUM_THREADS=16

SCRIPT_PATH=$(realpath $0)

link89

from streamlit.web.server.websocket_headers import _get_websocket_headers
from streamlit.components.v1 import html
import streamlit as st

from http.cookies import SimpleCookie
from uuid import uuid4
from time import sleep


def get_cookie():

link89

SCOW 数据表修改

通过数据库操作支持一些目前暂不支持在界面操作的 SCOW 功能。 以下操作需在 mis 的 db 容器中完成。

docker exec -it <container-id> bash
mysql -p 
<enter password>

link89

Summary

• build a new deepmd-kit plugin for an existed LAMMPS
• export new LD_LIBRARY_PATH and LAMMPS_PLUGIN_PATH properly

Build

bash build.sh

Test

link89

Note

• It's for CentOS 7, so the conda version have to be limited to cuda 11.8

Steps

CONDA_OVERRIDE_CUDA=11.8 nohup conda create -y -p ./3.0.0b0-cuda118 pytorch=2.1.2=cuda118*  python=3.11.* cudatoolkit=11.8  zlib cudnn cusparselt -c conda-forge --override-channels &
cd ./3.0.0b0-cuda118
source activate ./
pip install tensorflow==2.14.1
ln -s <path to source> source

link89

Summary

• Use conda to install pytorch will have less trouble than with pip
• Install lammps with openmpi support from conda-forge channel (or you can build your own lammps if you like)
• Install lammps from source or else your conda environemnt will be messed up!
• Install tensorflow with pip instead of conda as it is slow to resolve environment!
• Install pytorch and tensorflow with conda at the same time to avoid abi incompatible issue!
• For the sake of C++ABI compatibility, use conda to install tensorflow and pip for pytorch
• Don't use full oneapi with openmpi, just use mkl or else there will be segmentation fault

Steps

link89

Use copy to backup files

rclone copy ikkem-hpc-fq:/public/home/fqgong666/pretrained-model/all-test/dataset-2nd/2nd_data office:/public/dynacat/2nd_data --check-first  --verbose --dry-run

FQA

400 error of minio

If you see the following error

link89

Summary

• Use rocky8 singularity container to build and run to work around legacy glibc issue.

Build

 singularity exec -B /public,/etc --nv /public/groups/ai4ec/libs/images/rocky8.sif /bin/bash -l build_lammps.sh

Run