Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import DistilBertTokenizer, DistilBertModel
|
4 |
+
|
5 |
+
class SimilarityPredictor:
|
6 |
+
def __init__(self):
|
7 |
+
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
8 |
+
self.model = DistilBertModel.from_pretrained('patent_similarity_model').to(self.device)
|
9 |
+
self.tokenizer = DistilBertTokenizer.from_pretrained('patent_similarity_model')
|
10 |
+
self.head = torch.nn.Sequential(torch.nn.Linear(768, 1), torch.nn.Sigmoid()).to(self.device)
|
11 |
+
self.head.load_state_dict(torch.load('patent_similarity_model/head.pt', map_location=self.device))
|
12 |
+
|
13 |
+
def predict(self, anchor, target):
|
14 |
+
self.model.eval()
|
15 |
+
with torch.no_grad():
|
16 |
+
encoded = self.tokenizer(
|
17 |
+
[anchor],
|
18 |
+
[target],
|
19 |
+
padding=True,
|
20 |
+
truncation=True,
|
21 |
+
max_length=64,
|
22 |
+
return_tensors='pt'
|
23 |
+
).to(self.device)
|
24 |
+
|
25 |
+
output = self.head(self.model(**encoded)[0][:,0,:]).squeeze()
|
26 |
+
return float(output)
|
27 |
+
|
28 |
+
predictor = SimilarityPredictor()
|
29 |
+
|
30 |
+
# Örnek seçenekler
|
31 |
+
example_pairs = [
|
32 |
+
["mobile phone", "cellphone"],
|
33 |
+
["artificial intelligence", "machine learning"],
|
34 |
+
["electric vehicle", "battery powered car"],
|
35 |
+
["wireless communication", "radio transmission"],
|
36 |
+
["solar panel", "photovoltaic cell"],
|
37 |
+
["computer processor", "CPU"],
|
38 |
+
["digital storage", "memory device"],
|
39 |
+
["touch screen", "interactive display"],
|
40 |
+
["biometric authentication", "fingerprint recognition"],
|
41 |
+
["cloud computing", "remote server processing"]
|
42 |
+
]
|
43 |
+
|
44 |
+
def predict_similarity(anchor, target):
|
45 |
+
score = predictor.predict(anchor, target)
|
46 |
+
return round(score, 3)
|
47 |
+
|
48 |
+
# Create Gradio interface with examples
|
49 |
+
iface = gr.Interface(
|
50 |
+
fn=predict_similarity,
|
51 |
+
inputs=[
|
52 |
+
gr.Textbox(label="Anchor Phrase", placeholder="Enter first phrase..."),
|
53 |
+
gr.Textbox(label="Target Phrase", placeholder="Enter second phrase...")
|
54 |
+
],
|
55 |
+
outputs=gr.Number(label="Similarity Score (0-1)"),
|
56 |
+
title="Patent Phrase Similarity Checker",
|
57 |
+
description="""Compare the similarity between two patent phrases.
|
58 |
+
|
59 |
+
Score guide:
|
60 |
+
- 1.0: Very close match (exact or near-exact)
|
61 |
+
- 0.75: Close synonyms
|
62 |
+
- 0.5: Related terms
|
63 |
+
- 0.25: Somewhat related
|
64 |
+
- 0.0: Unrelated
|
65 |
+
|
66 |
+
Try the examples below or enter your own phrases!""",
|
67 |
+
examples=example_pairs,
|
68 |
+
theme="huggingface",
|
69 |
+
css="footer {display: none !important;}"
|
70 |
+
)
|
71 |
+
|
72 |
+
iface.launch()
|