Spaces:
Sleeping
Sleeping
init commit
Browse files- app.py +107 -0
- requirements.txt +5 -0
app.py
ADDED
@@ -0,0 +1,107 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
4 |
+
import numpy as np
|
5 |
+
import plotly.graph_objects as go
|
6 |
+
|
7 |
+
# Initialize model and tokenizer
|
8 |
+
MODEL_OPTIONS = {
|
9 |
+
"waleko/roberta-arxiv-tags": "RoBERTa Arxiv Tags"
|
10 |
+
}
|
11 |
+
|
12 |
+
def load_model(model_name):
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
14 |
+
model = AutoModelForSequenceClassification.from_pretrained(model_name)
|
15 |
+
return model, tokenizer
|
16 |
+
|
17 |
+
current_model = None
|
18 |
+
current_tokenizer = None
|
19 |
+
|
20 |
+
def get_model_and_tokenizer(model_name):
|
21 |
+
global current_model, current_tokenizer
|
22 |
+
if current_model is None or current_tokenizer is None:
|
23 |
+
current_model, current_tokenizer = load_model(model_name)
|
24 |
+
return current_model, current_tokenizer
|
25 |
+
|
26 |
+
def create_visualization(probs, labels):
|
27 |
+
return go.Figure(data=[go.Pie(
|
28 |
+
labels=labels + ['Others'] if sum(probs) < 1 else labels,
|
29 |
+
values=list(probs) + [1 - sum(probs)] if sum(probs) < 1 else list(probs),
|
30 |
+
textinfo='percent',
|
31 |
+
textposition='inside',
|
32 |
+
hole=.3,
|
33 |
+
showlegend=True
|
34 |
+
)])
|
35 |
+
|
36 |
+
def classify_text(title, abstract, model_name):
|
37 |
+
if not title and not abstract:
|
38 |
+
return "Error: At least one of title or abstract must be provided.", None
|
39 |
+
|
40 |
+
model, tokenizer = get_model_and_tokenizer(model_name)
|
41 |
+
text = 'Title: ' + (title or '') + '\n\nAbstract: ' + (abstract or '')
|
42 |
+
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=1024)
|
43 |
+
with torch.no_grad():
|
44 |
+
outputs = model(**inputs)
|
45 |
+
logits = outputs.logits
|
46 |
+
probs = torch.nn.functional.softmax(logits[0], dim=0)
|
47 |
+
probs = probs.numpy()
|
48 |
+
sorted_idx = np.argsort(probs)[::-1]
|
49 |
+
sorted_probs = probs[sorted_idx]
|
50 |
+
cumsum = np.cumsum(sorted_probs)
|
51 |
+
k = 1
|
52 |
+
if sorted_probs[0] < 0.95:
|
53 |
+
k = np.argmax(cumsum >= 0.95) + 1
|
54 |
+
id2label = model.config.id2label
|
55 |
+
tags = [id2label[idx] for idx in sorted_idx[:k]]
|
56 |
+
compact_pred = f'<span style="font-weight: 800;">{tags[0]}</span>' + (f" {' '.join(tags[1:])}" if len(tags) > 1 else "")
|
57 |
+
viz_data = create_visualization(
|
58 |
+
sorted_probs[:k],
|
59 |
+
[id2label[idx] for idx in sorted_idx[:k]]
|
60 |
+
)
|
61 |
+
html_output = f"""
|
62 |
+
<div>
|
63 |
+
<h3>Predicted Tags</h3>
|
64 |
+
<p>{compact_pred}</p>
|
65 |
+
</div>
|
66 |
+
"""
|
67 |
+
return html_output, viz_data
|
68 |
+
|
69 |
+
# Create Gradio interface
|
70 |
+
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
71 |
+
gr.Markdown("""
|
72 |
+
# Arxiv Tags Classification
|
73 |
+
Classify academic papers into arXiv categories using state-of-the-art language models.
|
74 |
+
""")
|
75 |
+
|
76 |
+
with gr.Row():
|
77 |
+
with gr.Column(scale=1):
|
78 |
+
model_dropdown = gr.Dropdown(
|
79 |
+
choices=list(MODEL_OPTIONS.keys()),
|
80 |
+
value=list(MODEL_OPTIONS.keys())[0],
|
81 |
+
label="Select Model",
|
82 |
+
info="Choose the model for classification"
|
83 |
+
)
|
84 |
+
title_input = gr.Textbox(
|
85 |
+
lines=1,
|
86 |
+
label="Title",
|
87 |
+
placeholder="Enter paper title (optional if abstract is provided)"
|
88 |
+
)
|
89 |
+
abstract_input = gr.Textbox(
|
90 |
+
lines=5,
|
91 |
+
label="Abstract",
|
92 |
+
placeholder="Enter paper abstract (optional if title is provided)"
|
93 |
+
)
|
94 |
+
with gr.Column(scale=1):
|
95 |
+
output_html = gr.HTML(
|
96 |
+
label="Predicted Tags"
|
97 |
+
)
|
98 |
+
output_plot = gr.Plot(
|
99 |
+
label="Probability Distribution",
|
100 |
+
show_label=True
|
101 |
+
)
|
102 |
+
inputs = [title_input, abstract_input, model_dropdown]
|
103 |
+
btn = gr.Button("Classify", variant="primary")
|
104 |
+
btn.click(fn=classify_text, inputs=inputs, outputs=[output_html, output_plot])
|
105 |
+
|
106 |
+
if __name__ == "__main__":
|
107 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio>=4.0.0
|
2 |
+
torch>=2.0.0
|
3 |
+
transformers>=4.30.0
|
4 |
+
numpy>=1.24.0
|
5 |
+
plotly
|