forked from lenscloth/GRPE
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpcqm4m_v1.sh
14 lines (13 loc) · 893 Bytes
/
pcqm4m_v1.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# PCQM4M Standard Model
python gpp.py --dataset lsc-v1 --world-size 8 --batch-size 128 \
--node-dim 768 --ffn-dim 768 --nhead 32 --num-layer 12 \
--warmup-epoch 3 --max-epoch 400 --attention-dropout 0.1 --dropout 0.1 \
--weight-decay 0.0 --peak-lr 2e-4 --max-hop 5 \
--valid-after 200 --valid-every 10 --save result/lsc_v1_large_model
# Evaluate Pretrained PCQM4M Standard Model
python gpp.py --dataset lsc-v1 --world-size 8 --batch-size 128 \
--node-dim 768 --ffn-dim 768 --nhead 32 --num-layer 12 \
--warmup-epoch 3 --max-epoch 400 --attention-dropout 0.1 --dropout 0.1 \
--weight-decay 0.0 --peak-lr 2e-4 --max-hop 5 \
--valid-after 200 --valid-every 10 --save result/lsc_v1_large_model_retrain \
--load pretrained_weight/pcqm4m_pretrained_standard.pt --load-all