Update AutoGluon max_memory from 0.1 to 0.4 in persist_models
#543
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
I've also opened a PR for AutoGluon to officially use 0.4 going forward as the default (although 0.8 still uses 0.1 as default) |
|
Merging this now, but please revert the change (or make it conditional), when a new release with the changed default is available. Thanks! |
PGijsbers
pushed a commit
that referenced
this pull request
Jun 22, 2023
PGijsbers
added a commit
that referenced
this pull request
Jun 22, 2023
* Update AutoGluon `max_memory` from 0.1 to 0.4 in persist_models (#543) * Add `optimize_for_deployment` for AutoGluon_hq (#544) * Reduce training time by 10% if a high_quality preset is used (#546) * Reduce training time by 10% if a high_quality preset is used High quality presets perform a post-fit step which takes 10~15% of total time (by Nick's estimate). To ensure comparisons stay reasonably fair we pre-emptively tell AutoGluon to use less time, so that all frameworks' models are based on "max_total_time" amount of effort. * Allow preset to be str or list and still reduce if hq or gq * Add identical markers to identify fit/inferencetime/predict stages (#548) * Add start_time, stop_time and log_time to failure.csv (#547) This helps more quickly identify at what stage the failure took place. E.g., if it's just a few minutes in, it is probably setup failure (such as connectivity issues). * Docker/permissions (#550) * Remove ownership changing and starting as user for docker images Since the USER is overwritten by `-u` for non-Windows platforms, which creates issues when the account running the docker image is not the same as the one that created it. * Dont run docker as root since images no longer have associated user * Ignore some additional files not needed to run the benchmark * Create root dir if it does not exist This is required, because otherwise in docker mode a non-existent directory is mounted, which is by default locked to `root` permissions. This in turn makes the benchmark app unable to create the subdirectories when the image is run as user. * Further remove user info from docker build and add run_as option The run_as option is then configurable so that it can be enabled for people who run into issues. Unfortunately, I observed different behavior from two systems with the same OS and docker versions installed. So for now I give up on one unified solution. * Update GAMA for v23.0.0 (#551) --------- Co-authored-by: Nick Erickson <innixma@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Updated max_memory to 0.4 as when tested it resulted in no crashes. The 0.1 default is overly conservative. For batch_size=1 in inference, this gives ~2x faster inference speed on average as more frequently models are able to fit into memory using this raised limit.