DepFAEnet-v3

• No tag information
• With tag 0 atoms
• Fine-tuned on selected adsorbates
• Ocdata

https://wandb.ai/mila-ocp/ocp-deup/runs/4648581-depfaenet-is2re-all

DepFAENet-v2

• No tag information
• With tag 0 atoms
• Fine-tuned on selected adsorbates

Wandb run:
https://wandb.ai/mila-ocp/ocp-alvaro/runs/4071859-depfaenet-is2re-all/overview?workspace=user-alex_duvalinho

DepFAENet not using tag info

• DepFAENet trained on whole dataset
• Without graph rewiring step (all atoms)
• Without tag information
• Not fine-tuned on specific adsorbates yet

Run:
https://wandb.ai/mila-ocp/ocp-alvaro/runs/4023244-depfaenet-is2re-all/overview?workspace=user-alex_duvalinho

Cleaned ICML submission version

What's Changed

• Simplify and unify model configuration files by @vict0rsch in #26
• Config refactor: improving and standardizing configs and args by @vict0rsch in #25
• [WIP] Energy gradient as target for the Forces by @vict0rsch in #27
• Handle QM9 dataset by @vict0rsch in #29
• Add notification mechanism by @vict0rsch in #28
• integrate QM9 dataset by @vict0rsch in #31
• Integrate QM7-X dataset by @vict0rsch in #32
• Gnn improvements by @AlexDuvalinho in #33
• Dev by @vict0rsch in #34
• Add line break in wandb_url.txt by @alexhernandezgarcia in #36
• Spherenet dig by @vict0rsch in #37
• Gnn improvements by @vict0rsch in #35
• ICML Clean-up by @vict0rsch in #38
• Dev by @vict0rsch in #30

Full Changelog: ai4mat-workshop...icml

What's Changed

• Simplify and unify model configuration files by @vict0rsch in #26
• Config refactor: improving and standardizing configs and args by @vict0rsch in #25
• [WIP] Energy gradient as target for the Forces by @vict0rsch in #27
• Handle QM9 dataset by @vict0rsch in #29
• Add notification mechanism by @vict0rsch in #28
• integrate QM9 dataset by @vict0rsch in #31
• Integrate QM7-X dataset by @vict0rsch in #32
• Gnn improvements by @AlexDuvalinho in #33
• Dev by @vict0rsch in #34
• Add line break in wandb_url.txt by @alexhernandezgarcia in #36
• Spherenet dig by @vict0rsch in #37
• Gnn improvements by @vict0rsch in #35
• ICML Clean-up by @vict0rsch in #38
• Dev by @vict0rsch in #30

Full Changelog: ai4mat-workshop...icml-prelease

ai4mat-workshop

What's Changed

• Adapt download_data.py script to fit Mila's cluster by @vict0rsch in #1
• Alex's branch to improve GGNs by @vict0rsch in #3
• blackify all things by @vict0rsch in #4
• Continue training if same $SLURM_JOBID and has checkpoint by @vict0rsch in #7
• Wandb-notes by @vict0rsch in #9
• Val logs by @AlexDuvalinho in #6
• Distributed training by @vict0rsch in #11
• Update GFlowNet scripts by @alexhernandezgarcia in #8
• line length -> 88 by @vict0rsch in #13
• black . with line length 88 by @alexhernandezgarcia in #14
• Fixed embeddings by @AlexDuvalinho in #10
• Fix DDP final eval by @vict0rsch in #16
• WandB Sweeps to run sbatch.py by @vict0rsch in #21
• Allow to use virtualenv environment instead of only conda in sbatch.py by @alexhernandezgarcia in #23
• [WIP] Super nodes by @vict0rsch in #17
• Time validation loop & some bug fixes by @vict0rsch in #22
• New gnn by @AlexDuvalinho in #19

New Contributors

• @AlexDuvalinho made their first contribution in #6
• @alexhernandezgarcia made their first contribution in #8

Full Changelog: https://github.com/RolnickLab/ocp/commits/ai4mat-workshop