jaeyong2 commited on
Commit
7f12fd5
1 Parent(s): 8ebec20

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +4 -5
README.md CHANGED
@@ -37,7 +37,7 @@ optimizer = AdamW(model.parameters(), lr=5e-5)
37
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
38
  model.to(device)
39
 
40
- for epoch in range(3): # 에포크 반복
41
  model.train()
42
  total_loss = 0
43
  count = 0
@@ -53,11 +53,11 @@ for epoch in range(3): # 에포크 반복
53
  positive_encodings = batch_to_device(positive_encodings, device)
54
  negative_encodings = batch_to_device(negative_encodings, device)
55
 
56
- # 모델 출력 (임베딩 벡터 생성)
57
- anchor_output = model(**anchor_encodings)[0][:, 0, :] # [CLS] 토큰의 벡터
58
  positive_output = model(**positive_encodings)[0][:, 0, :]
59
  negative_output = model(**negative_encodings)[0][:, 0, :]
60
- # 삼중항 손실 계산
61
  if loss==None:
62
  loss = triplet_loss(anchor_output, positive_output, negative_output)
63
  else:
@@ -91,7 +91,6 @@ def evaluate(validation_dataset):
91
  negative_embedding = get_embedding(item["Fake Title"], model, tokenizer)
92
 
93
 
94
- # 쿼리와 모든 문서 간의 유사도 계산 (코사인 거리 사용)
95
  positive_distances = pairwise_distances(query_embedding.detach().cpu().float().numpy(), document_embedding.detach().cpu().float().numpy(), metric="cosine")
96
  negative_distances = pairwise_distances(query_embedding.detach().cpu().float().numpy(), negative_embedding.detach().cpu().float().numpy(), metric="cosine")
97
 
 
37
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
38
  model.to(device)
39
 
40
+ for epoch in range(3):
41
  model.train()
42
  total_loss = 0
43
  count = 0
 
53
  positive_encodings = batch_to_device(positive_encodings, device)
54
  negative_encodings = batch_to_device(negative_encodings, device)
55
 
56
+
57
+ anchor_output = model(**anchor_encodings)[0][:, 0, :]
58
  positive_output = model(**positive_encodings)[0][:, 0, :]
59
  negative_output = model(**negative_encodings)[0][:, 0, :]
60
+
61
  if loss==None:
62
  loss = triplet_loss(anchor_output, positive_output, negative_output)
63
  else:
 
91
  negative_embedding = get_embedding(item["Fake Title"], model, tokenizer)
92
 
93
 
 
94
  positive_distances = pairwise_distances(query_embedding.detach().cpu().float().numpy(), document_embedding.detach().cpu().float().numpy(), metric="cosine")
95
  negative_distances = pairwise_distances(query_embedding.detach().cpu().float().numpy(), negative_embedding.detach().cpu().float().numpy(), metric="cosine")
96