config_hoip.yaml

# fold_num: 0
# dataset_name: 'matbench_log_gvrh'

# train_data: '/home/hongshuh/space_group_transformer/data/HOIP_train.csv'
# test_data: '/home/hongshuh/space_group_transformer/data/HOIP_test.csv'
train_data: '/home/hongshuh/space_group_transformer/data/MOF_train.csv'
test_data: '/home/hongshuh/space_group_transformer/data/MOF_test.csv'
val_data: '/home/hongshuh/space_group_transformer/data/MOF_val.csv'
target: 'CH4_0.05'
seed: 0
prompt: True
pretrain_model: null #'/home/hongshuh/space_group_transformer/runs_contrast/May19_09-50-37/best.pth'
# pretrain_model: null
vocab_path: '/home/hongshuh/space_group_transformer/mof_pore_vocab.json'
device: 'cuda:1'
epochs: 200
log: False
batch_size: 128
lr: 0.00001
weight_decay: 0.01
warmup_ratio: 0.05
robust: False
task: 'regression'
max_element: 20
hidden_size: 768
max_position_embeddings: 40            # max position embeddings of Transformer
blocksize: 38                         # max length of sequences after tokenization
num_attention_heads: 12                 # number of attention heads in each hidden layer
num_hidden_layers: 8                   # number of hidden layers
hidden_dropout_prob: 0.1                # hidden layer dropout
attention_probs_dropout_prob: 0.1       # attention dropout
mlm_probability: 0.15                   # masked probability in mlm