Manual for fine-tuning on Open Catalyst Project (OCP)

Based on fair-chem and its tutorial website (before Open Catalyst Project)

https://fair-chem.github.io/index.html

---

Welcome.

You might be able to fine-tune OCP models with your dataset after reading this manual.

Good luck.

---

A tiny description before you start

Entire fine-tuning process can be devided into two parts.

• Set a conda environment
• Fine-tuning

(There is an utilization part on bottom)

Much easier than running a DFT calculation, so you don't need to worry.

I shall be with you in every step.

---

All code is executed on our server.

-Linux CentOS release 7.9.2009

Mac OS X compatible with ase.db

Window not checked

---

Set a conda environment

The environment setup for Meta lab differs slightly from the OCP recommended setup.

I highly recommend creating a new environment using the 'env.gpu.yml' file.

Follow me.

1. Login to server

ssh, MobaXTerm, or anything you use

2. Download files through this github code

Type this on linux console

git clone https://github.com/GuGroup/Finetuner_OCP

3. Navigate to the Finetuner_OCP folder

cd Finetuner_OCP

4. Create a new environment with the following code

conda env create -f env.gpu.yml

(This process takes a few minutes.)

Tip

You can change the name of the environment via 'vi env.gpu.yml'

5. Activate git-chem

conda activate git-chem

6. Download fairchem-core

pip install fairchem-core==1.0.0

Now you are ready for fine-tuning!

---

Fine-tuning

One Step Fine-tuning

You can fine-tune with ytk's parameter pick.

Copy your ase file (OUTCAR, *.extxyz or *.traj) to Finetuner_OCP folder

cp OUTCAR /home/[your name]/Finetuner_OCP

Navigate to the 'Finetuner_OCP' folder

cd ~/Finetuner_OCP

ls

OUTCAR README.md dftocp.py env.gpu.yml finetuner.py main.py

Type this line on linux console

python finetuner.py --name [name_that_you_want] --path ./OUTCAR --db lmdb

Note

If name of your ase file is not OUTCAR, so like ytk.extxyz or gu.traj, change line like this

python finetuner.py --name ytk --path ./ytk.extxyz --db lmdb

First, This will be displayed on your console

Now making DB files in data folder...
100%|███████████████████████████████████████████████████████████████████████████████████| 73/73 [00:02<00:00, 30.13it/s]

It converts your ase file into lmdb (or asedb) and split into train, test, and val as 80:10:10 ratio, in data folder

Then you can see these lines

**Available S2EF models:
(IS2RE not supported)
CGCNN-S2EF-OC20-200k
CGCNN-S2EF-OC20-2M
CGCNN-S2EF-OC20-20M
CGCNN-S2EF-OC20-All
DimeNet-S2EF-OC20-200k
DimeNet-S2EF-OC20-2M
SchNet-S2EF-OC20-200k
SchNet-S2EF-OC20-2M
SchNet-S2EF-OC20-20M
SchNet-S2EF-OC20-All
DimeNet++-S2EF-OC20-200k
DimeNet++-S2EF-OC20-2M
DimeNet++-S2EF-OC20-20M
DimeNet++-S2EF-OC20-All
SpinConv-S2EF-OC20-2M
SpinConv-S2EF-OC20-All
GemNet-dT-S2EF-OC20-2M
GemNet-dT-S2EF-OC20-All
PaiNN-S2EF-OC20-All
GemNet-OC-S2EF-OC20-2M
GemNet-OC-S2EF-OC20-All
GemNet-OC-S2EF-OC20-All+MD
GemNet-OC-Large-S2EF-OC20-All+MD
SCN-S2EF-OC20-2M
SCN-t4-b2-S2EF-OC20-2M
SCN-S2EF-OC20-All+MD
eSCN-L4-M2-Lay12-S2EF-OC20-2M
eSCN-L6-M2-Lay12-S2EF-OC20-2M
eSCN-L6-M2-Lay12-S2EF-OC20-All+MD
eSCN-L6-M3-Lay20-S2EF-OC20-All+MD
EquiformerV2-83M-S2EF-OC20-2M
EquiformerV2-31M-S2EF-OC20-All+MD
EquiformerV2-153M-S2EF-OC20-All+MD
SchNet-S2EF-force-only-OC20-All
DimeNet++-Large-S2EF-force-only-OC20-All
DimeNet++-S2EF-force-only-OC20-20M+Rattled
DimeNet++-S2EF-force-only-OC20-20M+MD
GemNet-dT-S2EFS-OC22
GemNet-OC-S2EFS-OC22
GemNet-OC-S2EFS-OC20+OC22
GemNet-OC-S2EFS-nsn-OC20+OC22
GemNet-OC-S2EFS-OC20->OC22
EquiformerV2-lE4-lF100-S2EFS-OC22
SchNet-S2EF-ODAC
DimeNet++-S2EF-ODAC
PaiNN-S2EF-ODAC
GemNet-OC-S2EF-ODAC
eSCN-S2EF-ODAC
EquiformerV2-S2EF-ODAC
EquiformerV2-Large-S2EF-ODAC
**Type a model you want to Fine-tune

These are available pre-trained model names

In example of using GemNet-OC-S2EFS-OC20+OC22,

GemNet-OC-S2EFS-OC20+OC22

Pre-trained model is downloaded in your folder, as [pre-trained_model_name].pt

You might see some warnings.

WARNING:root:Detected old config, converting to new format. Consider updating to avoid potential incompatibilities.

WARNING:root:Unrecognized arguments: ['symmetric_edge_symmetrization']

WARNING:root:Using `weight_decay` from `optim` instead of `optim.optimizer_params`.Please update your config to use `optim.optimizer_params.weight_decay`.`optim.weight_decay` will soon be deprecated.

WARNING:root:No seed has been set in modelcheckpoint or OCPCalculator! Results may not be reproducible on re-run

Finally, fine-tuning is in progress

**Fine-tuning is in progress, check "train.txt"

In the same folder, train.txt is created

You can see how fine-tuning is going with this line.

tail -f -n 4 train.txt

---

Utilization

How to play with fine-tuned model?

Open ase.Atoms in python and put your fine-tuned model as calculator!

from farichem.core.common.relaxation.ase_utils import OCPCalculator
from ase.io import read

# Due to timestamping on folder name (2024-08-...), You need to change path of the checkpoint with your situation
checkpoint = './fine-tuning/checkpoints/2024-08-09-10-59-28-ytk/best_checkpoint.pt' 

# If you run this on a GPU node, cpu=False
newcalc = OCPCalculator(checkpoint_path=checkpoint, cpu=True)
traj = read('./OUTCAR', index=':')

atoms = traj[0]
atoms.calc = newcalc
print(atoms.get_potential_energy())

-118.8056945800

I uploaded 'dftocp.py' in Finetuner_OCP folder.

You can get .json file of potential energy in dft and ocp with it.

I recommend playing with it more..

---

This is the end...