Skip to content
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

fix optimizer save/load, add mpnn config, get f16 training to work #108

Merged
merged 7 commits into from
May 6, 2022

Conversation

jpata
Copy link
Owner

@jpata jpata commented May 5, 2022

  • Check that the MPNN option works (disabled by default)
  • fix optimizer state save/load (improve saving the optimizer state #109)
  • get f16 training to work by introducing pairwise_abs_distance (L1 norm), disabling normalize_degrees
  • add l1 distance norm for graph building (potentially numerically more stable)

@jpata jpata changed the title add mpnn config fix opt save/load, add mpnn config, get f16 training to work May 6, 2022
@jpata
Copy link
Owner Author

jpata commented May 6, 2022

Here's a comparison of the GCN training with f32 (red) and f16 (orange).

  • the loss is very similar
  • the training time is faster by almost 50%

Screenshot from 2022-05-06 13-15-29

@jpata jpata changed the title fix opt save/load, add mpnn config, get f16 training to work fix optimizer save/load, add mpnn config, get f16 training to work May 6, 2022
@jpata jpata merged commit e536b0f into master May 6, 2022
jpata added a commit that referenced this pull request Sep 15, 2023
)

* add mpnn config

* fix opt saving, f16

Former-commit-id: e536b0f
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant