Skip to content
This repository has been archived by the owner on Sep 18, 2024. It is now read-only.

Update config v2 doc #3711

Merged
merged 9 commits into from
Jun 8, 2021
Merged

Update config v2 doc #3711

merged 9 commits into from
Jun 8, 2021

Conversation

kvartet
Copy link
Contributor

@kvartet kvartet commented Jun 2, 2021

cannot find OpenPAI config reference link


3. Some fields take a path to file or directory.
Unless otherwise noted, both absolute path and relative path are supported, and ``~`` will be expanded to home directory.
.. _path:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

what is this .. _path: used for?

Copy link
Contributor Author

@kvartet kvartet Jun 3, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

An anchor, to show the meaning of the path. When quoted below, users can jump directly, like these:
Path_ to the JSON file containing the search space
Specify the :ref:`directory <path>` to place log, checkpoint, metadata, and other run-time stuff.

Experiment Config Reference
===========================
==============================
Experiment Config V2 Reference
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

suggest to remove "V2"

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

removed

@QuanluZhang QuanluZhang requested a review from liuzhe-lz June 3, 2021 03:22

A config file is needed when creating an experiment. The path of the config file is provided to ``nnictl``.
The config file is in YAML format.
This document describes the rules to write the config file, and provides some examples and templates.

.. Note:: This experiment config is still supported for now, but will be migrated to new `Experiment Config <../reference/experiment_config.rst>`_ in NNI v.
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@liuzhe-lz when not support config v1?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

there has been no plan to stop the support of v1


- `LocalConfig`_
- `RemoteConfig`_
- `OpenpaiConfig <openpai-class>`_
- :ref:`OpenpaiConfig <openpai-class>`
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why OpenpaiConfig is different?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.


`Mount point <path>`_ of storage service (typically NFS) on current machine.
:ref:`Mount point <path>` of storage service (typically NFS) on current machine.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

what is the rendering result of ":ref:Mount point <path>"?

Copy link
Contributor Author

@kvartet kvartet Jun 3, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

here #3711 (comment)
(..path)

@@ -638,15 +676,15 @@ default: ``False``


openpaiConfig
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Because OpenPai training service has a config field named openpaiConfig, and the headers are not case sensitive. Just use openpaiConfig can not link to the correct section, and old anchor is invalid, so I use :ref

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

got it, thx.


Specify whether NNI should submit trials to GPUs occupied by other tasks.

type: ``Optional[bool]``

Must be set when `trialGpuNumber` greater than zero.

If your are using desktop system with GUI, set this to ``True``.
If you are using the desktop system with GUI, set this to ``True``.
Copy link
Contributor

@liuzhe-lz liuzhe-lz Jun 3, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please add these detail info:

Following processes can make GPU "active":

  • non-NNI CUDA programs
  • graphical desktop
  • trials submitted by other NNI instances, if you have more than one NNI experiments running at same time
  • other users' CUDA programs, if you are using a shared server

Therefore, if you are using graphical OS like Windows 10 or Ubuntu desktop and useActiveGpu is set to False, the GUI will prevent NNI from launching any trial.
If you creates multiple NNI experiments and useActiveGpu is set to True, they will submit multiple trials to one GPU.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

... and useActiveGpu is set to False, the GUI will prevent NNI from launching any trial?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yay, they were flipped.

Copy link
Contributor Author

@kvartet kvartet Jun 3, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

some questions

  1. If I create multiple NNI experiments and useActiveGpu is set to True, only use one GPU, instead of related to gpuIndices?
  2. In OpenPAI and AML, when I want to use GPU, what should I do?
  3. the type of tunerGpuIndices and gpuIndices can be set to int?

Copy link
Contributor

@liuzhe-lz liuzhe-lz Jun 4, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. Assume there is only one GPU on the machine, then each experiment will submit one trial to that GPU. The GPU will have 2 trials run concurrently.
  2. Set trialGpuNumber, that's all. Generally you should not need fine-grained control like useActiveGpu on "managed" platforms.
  3. Yes, gpuIndices: 0,1,2 is string, gpuIndices: 1,2 is string, but gpuIndices: 2 is int.

@kvartet kvartet requested review from QuanluZhang and liuzhe-lz June 8, 2021 03:01
@ultmaster ultmaster merged commit d1b1e7b into microsoft:master Jun 8, 2021
@kvartet kvartet deleted the config-doc branch June 17, 2021 05:27
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants