Skip to content

Commit

Permalink
v0.0.3 release changes (#256)
Browse files Browse the repository at this point in the history
* Changes for v0.0.3

* Reformatting MODELS.md

* GemNet scaling factors

* Env update

* Correct pytorch version in readme

* Update OCPCalc docs

* gemnet scaling factors file

* remove unused var

* remove redundant lines in collate fn

* Updates gitignore

Co-authored-by: Muhammed Shuaibi <mshuaibi@andrew.cmu.edu>
  • Loading branch information
abhshkdz and mshuaibii committed Aug 27, 2021
1 parent 9812420 commit 65c2d62
Show file tree
Hide file tree
Showing 59 changed files with 6,009 additions and 643 deletions.
2 changes: 1 addition & 1 deletion .circleci/config.yml
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@ jobs:
conda activate ocp-models
pip install -e .
pre-commit install
pytest
pytest /home/circleci/project/tests
- run:
name: Run black
Expand Down
10 changes: 7 additions & 3 deletions .gitignore
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
wandb/
*data*
wandb
data
checkpoints
results
logs
*.traj
*.json
experimental

# Byte-compiled / optimized / DLL files
__pycache__/
Expand Down Expand Up @@ -70,6 +70,7 @@ instance/

# Sphinx documentation
docs/_build/
docs/source/_build/

# PyBuilder
target/
Expand Down Expand Up @@ -107,3 +108,6 @@ Local

# PyCharm
.idea/

# VS Code
.vscode/
3 changes: 2 additions & 1 deletion .pre-commit-config.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@ repos:
- id: check-added-large-files
- id: end-of-file-fixer
- repo: https://github.com/pre-commit/mirrors-isort
rev: v4.3.21
rev: v5.9.1
hooks:
- id: isort
args: ["--profile", "black", "--filter-files"]
8 changes: 4 additions & 4 deletions DATASET_PER_ADSORBATE.md
Original file line number Diff line number Diff line change
Expand Up @@ -84,13 +84,13 @@ Download links are in the table below:

Note - A few adsorbates are intentionally left out for the test splits.

Downloading any of the above and extracting will result in a folder :
Downloading any of the above and extracting will result in a folder:

`<index>/`

* `system.txt` Text file containing information about the different adsorbate+catalyst system names. In total there are N systems. More details described below.
* `<index>/`
* This contains N compressed trajectory files of the format `.extxyz.xz`.
* `<index>/`
* This contains N compressed trajectory files of the format `.extxyz.xz`.
* Files are named as `<system_id>.extxyz.xz` (where `system_id` is defined below).


Expand All @@ -107,6 +107,6 @@ where:
* `reference_energy` - Energy used to reference system energies to bare catalyst+gas reference energies. Used for adsorption energy calculations.


The `.extxyz.xz` files are LZMA compressed `.extxyz` trajectory files. Each trajectory corresponds to a relaxation trajectory of a different adsorbate+catalyst system. Information about the `.extxyz` trajectory file format may be found at https://wiki.fysik.dtu.dk/ase/dev/ase/io/formatoptions.html#extxyz .
The `.extxyz.xz` files are LZMA compressed `.extxyz` trajectory files. Each trajectory corresponds to a relaxation trajectory of a different adsorbate+catalyst system. Information about the `.extxyz` trajectory file format may be found at https://wiki.fysik.dtu.dk/ase/ase/io/formatoptions.html#extxyz.

In order to uncompress the files, `uncompress.py` provides a multi-core implementation which could be used.
102 changes: 47 additions & 55 deletions MODELS.md
Original file line number Diff line number Diff line change
@@ -1,71 +1,63 @@
# Pretrained models for OCP

This page summarizes all the pretrained models released as part of the [Open Catalyst Project](https://opencatalystproject.org/). All models were trained using this codebase in October 2020.
This page summarizes all the pretrained models released as part of the [Open Catalyst Project](https://opencatalystproject.org/). All models were trained using this codebase.

* All configurations for these baseline models are available in the [`configs/`](https://github.com/Open-Catalyst-Project/ocp/tree/master/configs) directory.
* All configurations for these models are available in the [`configs/`](https://github.com/Open-Catalyst-Project/ocp/tree/master/configs) directory.
* All of these models are trained on various splits of the OC20 S2EF / IS2RE datasets. For details, see [https://arxiv.org/abs/2010.09990v1](https://arxiv.org/abs/2010.09990) and https://github.com/Open-Catalyst-Project/ocp/blob/master/DATASET.md.
* All model checkpoints were created using Pytorch 1.6. Please follow steps listed in the [readme](https://github.com/open-catalyst-Project/ocp#installation) to set up your conda environment with correct package versions.

## S2EF baselines: optimized for EFwT



|model |split |downloadable link |val ID force MAE |val ID EFwT |
## S2EF models: optimized for EFwT

|model |split |input path on FAIR cluster |downloadable link |val ID force MAE |val ID EFwT |
|--- |--- |--- |--- |--- |--- |
|CGCNN |200k |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/cgcnn_200k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_200k.pt |0.08 |0% |
|CGCNN |2M |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/cgcnn_2M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_2M.pt |0.0673 |0.01% |
|CGCNN |20M |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/cgcnn_20M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_20M.pt |0.065 |0% |
|CGCNN |All |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/cgcnn_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_all.pt |0.0684 |0.01% |
|DimeNet |200k |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/dimenet_200k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/dimenet_200k.pt |0.0693 |0.01% |
|DimeNet |2M |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/dimenet_2M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/dimenet_2M.pt |0.0576 |0.02% |
|SchNet |200k |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/schnet_200k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_200k.pt |0.0743 |0% |
|SchNet |2M |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/schnet_2M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_2M.pt |0.0737 |0% |
|SchNet |20M |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/schnet_20M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_20M.pt |0.0568 |0.03% |
|SchNet |All |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/schnet_all_large.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_all_large.pt |0.0494 |0.12% |
|DimeNet++ |200k |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_200k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_200k.pt |0.0741 |0% |
|DimeNet++ |2M |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_2M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_2M.pt |0.0595 |0.01% |
|DimeNet++ |20M |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_20M.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_20M.pt |0.0511 |0.06% |
|DimeNet++ |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_all.pt |0.0444 |0.12% |
|SpinConv |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/08_26_ocp_checkpoints_share/s2ef/spinconv_force_centric_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_08/s2ef/spinconv_force_centric_all.pt |0.0267 |1.02% |
|GemNet-dT |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/08_26_ocp_checkpoints_share/s2ef/gemnet_t_direct_h512_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_08/s2ef/gemnet_t_direct_h512_all.pt |0.0211 |2.21% |

## S2EF models: optimized for force only

|model |split |input path on FAIR cluster |downloadable link |val ID force MAE |
|--- |--- |--- |--- |--- |
|CGCNN |200k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_200k.pt |0.08 |0% |
|CGCNN |2M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_2M.pt |0.0673 |0.01% |
|CGCNN |20M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_20M.pt |0.065 |0% |
|CGCNN |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/cgcnn_all.pt |0.0684 |0.01% |
|DimeNet |200k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/dimenet_200k.pt |0.0693 |0.01% |
|DimeNet |2M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/dimenet_2M.pt |0.0576 |0.02% |
|SchNet |200k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_200k.pt |0.0743 |0% |
|SchNet |2M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_2M.pt |0.0737 |0% |
|SchNet |20M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_20M.pt |0.0568 |0.03% |
|SchNet |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_all_large.pt |0.0494 |0.12% |
|DimeNet++ |200k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_200k.pt |0.0741 |0% |
|DimeNet++ |2M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_2M.pt |0.0595 |0.01% |
|DimeNet++ |20M |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_20M.pt |0.0511 |0.06% |
|DimeNet++ |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_all.pt |0.0444 |0.12% |

## S2EF baselines: optimized for force only

##

|model |split |downloadable link |val ID force MAE |
|--- |--- |--- |--- |
|SchNet |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_all_forceonly.pt |0.0443 |
|DimeNet++ |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/dimenetpp_all_forceonly.pt |0.0334 |
|DimeNet++-Large |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_large_all_forceonly.pt |0.02825 |
|DimeNet++ |20M+Rattled |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_20M_rattled_forceonly.pt |0.0614 |
|DimeNet++ |20M+MD |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_20M_md_forceonly.pt |0.0594 |

|SchNet |All |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/schnet_all_forceonly.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/schnet_all_forceonly.pt |0.0443 |
|DimeNet++ |All |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/s2ef/dimenetpp_all_forceonly.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/s2ef/dimenetpp_all_forceonly.pt |0.0334 |
|DimeNet++-Large |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_large_all_forceonly.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_large_all_forceonly.pt |0.02825 |
|DimeNet++ |20M+Rattled |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_20M_rattled_forceonly.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_20M_rattled_forceonly.pt |0.0614 |
|DimeNet++ |20M+MD |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/s2ef/dimenetpp_20M_md_forceonly.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/s2ef/dimenetpp_20M_md_forceonly.pt |0.0594 |

## IS2RE models

## IS2RE baselines



|model |split |downloadable link |val ID energy MAE |
|--- |--- |--- |--- |
|CGCNN |10k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/cgcnn_10k.pt |0.9881 |
|CGCNN |100k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/cgcnn_100k.pt |0.6820 |
|CGCNN |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/cgcnn_all.pt |0.6199 |
|DimeNet |10k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/is2re/dimenet_10k.pt |1.0117 |
|DimeNet |100k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/is2re/dimenet_100k.pt |0.6658 |
|DimeNet |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/is2re/dimenet_all.pt |0.5999 |
|SchNet |10k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/schnet_10k.pt |1.0590 |
|SchNet |100k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/schnet_100k.pt |0.7137 |
|SchNet |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/schnet_all.pt |0.6458 |
|DimeNet++ |10k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/dimenetpp_10k.pt |0.8837 |
|DimeNet++ |100k |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/dimenetpp_100k.pt |0.6388 |
|DimeNet++ |All |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/dimenetpp_all.pt |0.5639 |
|model |split |input path on fair cluster |downloadable link |val ID energy MAE |
|--- |--- |--- |--- |--- |
|CGCNN |10k |c |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/cgcnn_10k.pt |0.9881 |
|CGCNN |100k |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/cgcnn_100k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/cgcnn_100k.pt |0.682 |
|CGCNN |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/cgcnn_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/cgcnn_all.pt |0.6199 |
|DimeNet |10k |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/is2re/dimenet_10k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/is2re/dimenet_10k.pt |1.0117 |
|DimeNet |100k |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/is2re/dimenet_100k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/is2re/dimenet_100k.pt |0.6658 |
|DimeNet |All |/checkpoint/abhshkdz/10_19_ocp_checkpoints_share/is2re/dimenet_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2020_11/is2re/dimenet_all.pt |0.5999 |
|SchNet |10k |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/schnet_10k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/schnet_10k.pt |1.059 |
|SchNet |100k |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/schnet_100k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/schnet_100k.pt |0.7137 |
|SchNet |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/schnet_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/schnet_all.pt |0.6458 |
|DimeNet++ |10k |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/dimenetpp_10k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/dimenetpp_10k.pt |0.8837 |
|DimeNet++ |100k |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/dimenetpp_100k.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/dimenetpp_100k.pt |0.6388 |
|DimeNet++ |All |/checkpoint/abhshkdz/open-catalyst-project/checkpoints/02_22_ocp_checkpoints_share/is2re_from_muhammed/dimenetpp_all.pt |https://dl.fbaipublicfiles.com/opencatalystproject/models/2021_02/is2re/dimenetpp_all.pt |0.5639 |

The Open Catalyst 2020 (OC20) dataset is licensed under a [Creative Commons Attribution 4.0 License](https://creativecommons.org/licenses/by/4.0/legalcode).

Please cite the following paper in any research manuscript using the OC20 dataset or pretrained models:


```bibtex
```
@article{ocp_dataset,
author = {Chanussot*, Lowik and Das*, Abhishek and Goyal*, Siddharth and Lavril*, Thibaut and Shuaibi*, Muhammed and Riviere, Morgane and Tran, Kevin and Heras-Domingo, Javier and Ho, Caleb and Hu, Weihua and Palizhati, Aini and Sriram, Anuroop and Wood, Brandon and Yoon, Junwoong and Parikh, Devi and Zitnick, C. Lawrence and Ulissi, Zachary},
title = {Open Catalyst 2020 (OC20) Dataset and Community Challenges},
Expand Down
2 changes: 2 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,8 @@ ocp-models is the modeling codebase for the [Open Catalyst Project](https://open
It provides implementations of state-of-the-art ML algorithms for catalysis that
take arbitrary chemical structures as input to predict energy / forces / positions:

- [GemNet](https://arxiv.org/abs/2106.08903)
- [SpinConv](https://arxiv.org/abs/2106.09575)
- [DimeNet++](https://arxiv.org/abs/2011.14115)
- [ForceNet](https://arxiv.org/abs/2103.01436)
- [DimeNet](https://arxiv.org/abs/2003.03123)
Expand Down
55 changes: 55 additions & 0 deletions configs/s2ef/200k/gemnet/gemnet-dT.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
includes:
- configs/s2ef/200k/base.yml

model:
name: gemnet_t
num_spherical: 7
num_radial: 128
num_blocks: 3
emb_size_atom: 512
emb_size_edge: 512
emb_size_trip: 64
emb_size_rbf: 16
emb_size_cbf: 16
emb_size_bil_trip: 64
num_before_skip: 1
num_after_skip: 2
num_concat: 1
num_atom: 3
cutoff: 6.0
max_neighbors: 50
rbf:
name: gaussian
envelope:
name: polynomial
exponent: 5
cbf:
name: spherical_harmonics
extensive: True
otf_graph: False
output_init: HeOrthogonal
activation: silu
scale_file: configs/s2ef/all/gemnet/scaling_factors/gemnet-dT.json

regress_forces: True
direct_forces: True

optim:
batch_size: 32
eval_batch_size: 32
eval_every: 5000
num_workers: 2
lr_initial: 5.e-4
optimizer: AdamW
optimizer_params: {"amsgrad": True}
scheduler: ReduceLROnPlateau
mode: min
factor: 0.8
patience: 3
max_epochs: 80
force_coefficient: 100
energy_coefficient: 1
ema_decay: 0.999
clip_grad_norm: 10
loss_energy: mae
loss_force: l2mae
37 changes: 37 additions & 0 deletions configs/s2ef/200k/spinconv/spinconv_force.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
includes:
- configs/s2ef/200k/base.yml

model:
name: spinconv
model_ref_number: 0
hidden_channels: 32
mid_hidden_channels: 256
num_interactions: 3
num_basis_functions: 512
sphere_size_lat: 16
sphere_size_long: 12
max_num_neighbors: 40
cutoff: 6.0
sphere_message: fullconv
output_message: fullconv
force_estimator: random
regress_forces: True
use_pbc: True
scale_distances: True
basis_width_scalar: 3.0

optim:
batch_size: 3
eval_batch_size: 3
num_workers: 8
lr_initial: 0.0004
optimizer: Adam
optimizer_params: {"amsgrad": True}
eval_every: 5000
scheduler: ReduceLROnPlateau
mode: min
factor: 0.8
patience: 3
max_epochs: 80
force_coefficient: 100
energy_coefficient: 1
55 changes: 55 additions & 0 deletions configs/s2ef/20M/gemnet/gemnet-dT.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
includes:
- configs/s2ef/20M/base.yml

model:
name: gemnet_t
num_spherical: 7
num_radial: 128
num_blocks: 3
emb_size_atom: 512
emb_size_edge: 512
emb_size_trip: 64
emb_size_rbf: 16
emb_size_cbf: 16
emb_size_bil_trip: 64
num_before_skip: 1
num_after_skip: 2
num_concat: 1
num_atom: 3
cutoff: 6.0
max_neighbors: 50
rbf:
name: gaussian
envelope:
name: polynomial
exponent: 5
cbf:
name: spherical_harmonics
extensive: True
otf_graph: False
output_init: HeOrthogonal
activation: silu
scale_file: configs/s2ef/all/gemnet/scaling_factors/gemnet-dT.json

regress_forces: True
direct_forces: True

optim:
batch_size: 32
eval_batch_size: 32
eval_every: 5000
num_workers: 2
lr_initial: 5.e-4
optimizer: AdamW
optimizer_params: {"amsgrad": True}
scheduler: ReduceLROnPlateau
mode: min
factor: 0.8
patience: 3
max_epochs: 80
force_coefficient: 100
energy_coefficient: 1
ema_decay: 0.999
clip_grad_norm: 10
loss_energy: mae
loss_force: l2mae
37 changes: 37 additions & 0 deletions configs/s2ef/20M/spinconv/spinconv_force.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
includes:
- configs/s2ef/20M/base.yml

model:
name: spinconv
model_ref_number: 0
hidden_channels: 32
mid_hidden_channels: 256
num_interactions: 3
num_basis_functions: 512
sphere_size_lat: 16
sphere_size_long: 12
max_num_neighbors: 40
cutoff: 6.0
sphere_message: fullconv
output_message: fullconv
force_estimator: random
regress_forces: True
use_pbc: True
scale_distances: True
basis_width_scalar: 3.0

optim:
batch_size: 3
eval_batch_size: 3
num_workers: 8
lr_initial: 0.0004
optimizer: Adam
optimizer_params: {"amsgrad": True}
eval_every: 5000
scheduler: ReduceLROnPlateau
mode: min
factor: 0.8
patience: 3
max_epochs: 80
force_coefficient: 100
energy_coefficient: 1
Loading

0 comments on commit 65c2d62

Please sign in to comment.