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

link89

#!/bin/bash

trap 'echo ffmpegexit && pkill -P $$'  EXIT SIGINT SIGTERM SIGHUP

# Define the timeout duration in seconds for file change
TIMEOUT_IN_S=10
TOTAL_TIMEOUT_IN_S=7200

# Get the last argument as the output file
OUTPUT_FILE="${@: -1}"

link89

Background

• A windows XP guest machine exported from VMWare as OVF format

Tips

• Install cockpit and cockpick-machines: sudo apt install cockpit cockpit-machines
• Convert vmdk image to qcow2 with the following command: qemu-img convert -f vmdk -O qcow2 xp-wire-disk1.vmdk xp-wire-disk1.qcow2
• Create winxp virtual machine by importing existed disk, OS type should select unknown, don't start vm after import.
• Edit network interfaces, change model to e1000(PCI, legacy)
• Download e1000 driver iso from: https://archive.org/details/intel-pro1000-mt-desktop-82540em-winxp
• Add ISO as CD/ROM

link89

Docker

set proxy for docker build and docker run

You can add proxy configurations for the Docker client using a JSON configuration file, located in ~/.docker/config.json. Builds and containers use the configuration specified in this file.

{
 "proxies": {
   "default": {
     "httpProxy": "http://proxy.example.com:3128",
 "httpsProxy": "https://proxy.example.com:3129",

link89

2.2.8

CONDA_OVERRIDE_CUDA="11.8" conda create -p deepmd/2.2.8/gpu deepmd-kit=2.2.8=cuda118py310*  lammps horovod -c conda-forge 

To work around a known issue of conda-forge's tensorflow package, one has to run pip install nvidia-cuda-nvcc-cu11 in the conda environment.

Before 2.2.8

link89

"""
Convert DIYP TV List to m3u format

Usage:
python dipy_to_m3u.py diyp.txt out.m3u

DIPY format example:
cctv|channel,#genre#
CCTV1,http://example.org
"""

link89

// ==UserScript==
// @name         Reddit Crawler
// @namespace    http://tampermonkey.net/
// @version      0.1
// @description  Crawl posts and comments on a specific reddit sub and export as jsonl
// @author       Bing
// @match        https://www.reddit.com/*
// @grant        GM_registerMenuCommand
// ==/UserScript==

link89

Build DeepMD-kit + LAMMPS in a conda envrionment

Steps

Create conda environment

conda create -p 2.2.5/gpu python==3.10.13  tensorflow=2.10.*=gpu*

Note:

• Install tensorflow with conda instead of pip will have less trouble (for example, conda will install the right cudnn for you)