config=config).to(device)
except BaseException:
bert_model = torch.load(os.path.join(embedding,
'pytorch_model.bin')).to(device)
model = LinearModel(len(rel2id), embedding_type, bert_model,
freeze_embedding).to(device)
# optimizier
if embedding_type != 'bert':
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
if embedding_type == "bert":
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [
p for n, p in model.named_parameters()
if not any(nd in n for nd in no_decay)
],
"weight_decay":
0.0,
},
{
"params": [
p for n, p in model.named_parameters()
if any(nd in n for nd in no_decay)
],
"weight_decay":
0.0
},
]
optimizer = AdamW(optimizer_grouped_parameters, lr=learning_rate, eps=1e-8)
else:
try:
config = AutoConfig.from_pretrained(embedding)
bert_model = AutoModel.from_pretrained(embedding, config=config).to(device)
except BaseException:
bert_model = torch.load(os.path.join(embedding, 'pytorch_model.bin')).to(device)
model = LinearModel(len(rel2id), embedding_type, bert_model, freeze_embedding).to(device)
# optimizier
if embedding_type != 'bert':
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
if embedding_type == "bert":
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
"weight_decay": 0.0,
},
{"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
]
optimizer = AdamW(optimizer_grouped_parameters, lr=learning_rate, eps=1e-8)
scheduler = get_linear_schedule_with_warmup(optimizer,
num_warmup_steps=int(epoch_num * len(train_dataloader) * 0.1),
num_training_steps=epoch_num * len(train_dataloader))
# Prepare eval function
from sklearn.metrics import accuracy_score, classification_report, f1_score
def eval(m, dataloader):
y_pred = []
y_true = []
