Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -11,19 +11,20 @@ def get_bert_embeddings(words):
|
|
11 |
model = BertModel.from_pretrained('bert-base-uncased')
|
12 |
embeddings = []
|
13 |
|
14 |
-
# Extract embeddings
|
15 |
for word in words:
|
16 |
inputs = tokenizer(word, return_tensors='pt')
|
17 |
outputs = model(**inputs)
|
18 |
-
|
|
|
|
|
19 |
|
20 |
-
# Reduce dimensions to 3 using PCA
|
21 |
if len(embeddings) > 0:
|
22 |
pca = PCA(n_components=3)
|
23 |
reduced_embeddings = pca.fit_transform(np.array(embeddings))
|
24 |
return reduced_embeddings
|
25 |
return []
|
26 |
|
|
|
27 |
# Plotly plotting function
|
28 |
def plot_interactive_bert_embeddings(embeddings, words):
|
29 |
if len(words) < 4:
|
|
|
11 |
model = BertModel.from_pretrained('bert-base-uncased')
|
12 |
embeddings = []
|
13 |
|
|
|
14 |
for word in words:
|
15 |
inputs = tokenizer(word, return_tensors='pt')
|
16 |
outputs = model(**inputs)
|
17 |
+
# Use the [CLS] token's embedding
|
18 |
+
cls_embedding = outputs.last_hidden_state[0][0].detach().numpy()
|
19 |
+
embeddings.append(cls_embedding)
|
20 |
|
|
|
21 |
if len(embeddings) > 0:
|
22 |
pca = PCA(n_components=3)
|
23 |
reduced_embeddings = pca.fit_transform(np.array(embeddings))
|
24 |
return reduced_embeddings
|
25 |
return []
|
26 |
|
27 |
+
|
28 |
# Plotly plotting function
|
29 |
def plot_interactive_bert_embeddings(embeddings, words):
|
30 |
if len(words) < 4:
|