Skip to content

Enable custom perf models in MemoryBalancedPartitioner #2897

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from

Conversation

micrain
Copy link

@micrain micrain commented Apr 19, 2025

Summary:
We are enabling custom perf models while evaluating candidate plans in MemoryBalancedPartitioner. This will enable us to leverage the new critical path model defined in NoopCriticalPathPerfModel while trying to construct a memory balanced plan with the greedy partitioner.

Upcoming diffs will modify the greedy partitioner to leverage the new perf model.

Differential Revision: D73292939

Caner Gocmen added 2 commits April 18, 2025 17:35
Summary:

Update the critical path definition in the planner logs to match what we think is the most realistic option. See the docstring for the detailed modeling.

The goal is to later feed this into the proposer and partitioner to test if we can improve upon the current greedy algorithm used with the new perf model. 

We will also replace a [similar version](https://www.internalfb.com/code/fbsource/[50e47c413eb3d4e00facb9df71592f9aa81d8aee]/fbcode/torchrec/distributed/planner/stats.py?lines=1071) of this function in the stats.py to avoid duplicating the logic.

Differential Revision: D73207877
Summary:
We are enabling custom perf models while evaluating candidate plans in  `MemoryBalancedPartitioner`. This will enable us to leverage the new critical path model defined in `NoopCriticalPathPerfModel` while trying to construct a memory balanced plan with the greedy partitioner.

Upcoming diffs will modify the greedy partitioner to leverage the new perf model.

Differential Revision: D73292939
@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Apr 19, 2025
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D73292939

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. fb-exported
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants