Skip to content

Latest commit

 

History

History
113 lines (93 loc) · 4.58 KB

tuxiv.conf.md

File metadata and controls

113 lines (93 loc) · 4.58 KB

ENVIRONMENT VARIABLES

  • TACC_WORKDIR: TACC job workspace directory. Each job has a different workspace directory.
  • TACC_USERDIR: TACC User directory.
  • TACC_SLURM_USERLOG: Slurm log directory. The default value is ${TACC_USERDIR}/slurm_log.
  • tip: These ENVIRONMENT VARIABLES can be used in tuxiv.conf or read in python code.

There are four parts in tuxiv.conf that configure different parts of job submission. Noted that tuxiv.conf follows yaml format.

Entrypoint

In this section, you should input you shell commands to run your code line-by-line. The tcloud CLI will help run the job according to your commands.

entrypoint:
    - python ${TACC_WORKDIR}/mnist.py --epoch=3

Environment

In this section, you can specify your software requirements, including the environment name, dependencies, source channels and so on. The tcloud CLI will help build your environment with miniconda.

Notice: The environment name is optional. You can have the following two options.

  1. Environment name set.

    In this case, tcloud will create a new environment when you change any of your dependencies configuration.

    environment:
  dependencies:
      - pytorch=1.6.0
      - torchvision=0.7.0
  channels: pytorch

  2. Environment name unset.

    In this case, the environment will be persistent and tcloud will be updated the environment when you change any of your dependencies configuration (instead of creat a new environment). The environment configuration of tcloud is managed by conda, and you can follow conda to manage your environment.

    environment:
    name: torch-env
    dependencies:
        - pytorch=1.6.0
        - torchvision=0.7.0
  channels: pytorch

  • Check environment

    Check the existing environment with the tcloud env ls command.

    ~ ❯ tcloud env ls
# conda environments:
#
base                  *  /mnt/home/username/.Miniconda3
pytorch                  /mnt/home/username/.Miniconda3/envs/pytorch

    Check installed dependencies in a sp environment the existing environment dependencies with the tcloud env ls -n [ENV_NAME] command.

    ~ ❯ tcloud env ls -n base                                                                        
# packages in environment at /mnt/home/username/.Miniconda3:
#
# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main
brotlipy                  0.7.0           py38h27cfd23_1003
ca-certificates           2020.10.14                    0
certifi                   2020.6.20          pyhd3eb1b0_3
cffi                      1.14.3           py38h261ae71_2
chardet                   3.0.4           py38h06a4308_1003
conda                     4.9.2            py38h06a4308_0
conda-package-handling    1.7.2            py38h03888b9_0
...

Job

In this section, you can specify your configurations for cluster resources, including number of nodes, CPUs, GPUs, output file and so on. All the cluster configuration should be set in the general part.

job:
  name: test
  general:
    - nodes=2                # the number of nodes
    - ntasks-per-node=1      # the number of tasks per node
    - cpus-per-task=10       # the number of cpu per task
    - gres=gpu:2             # the number of gpu per node

Note: You can modify the output log path in Job section. For debugging purpose, we recommend you set the output value under ${TACC_USERDIR} directory and check it using tcloud ls and tcloud download.

Datasets

  • tcloud will help place the public datasets access in TACC_USERDIR. You can view the table of datasets at Dataset Info or check the table below.

    • Dataset Name
      0 imagenet
      1 mnist
      2 cifar-10
      3 coco17
      4 more datasets upon request

    • to access the public dataset you need to add this command in your tuxiv.conf file:

      datasets:
  - imagenet

    • also use this path as a dataset directory:

      ${TACC_USERDIR}/DATASET_NAME

  • User dataset: if you want to use your own dataset, you may skip this part and follow the instructions to upload and use your dataset.