Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,8 +1,7 @@
|
|
1 |
import gradio as gr
|
2 |
import requests
|
3 |
-
from typing import Dict, List, Tuple
|
4 |
|
5 |
-
def get_features(text: str)
|
6 |
url = "https://www.neuronpedia.org/api/search-with-topk"
|
7 |
payload = {
|
8 |
"modelId": "gemma-2-2b",
|
@@ -30,43 +29,70 @@ def create_dashboard(feature_id: int) -> str:
|
|
30 |
</div>
|
31 |
"""
|
32 |
|
33 |
-
def
|
34 |
if not text:
|
35 |
-
return gr.
|
36 |
-
|
37 |
features_data = get_features(text)
|
38 |
if not features_data:
|
39 |
-
return gr.
|
40 |
-
|
|
|
41 |
first_feature_id = None
|
42 |
-
with gr.Column() as col:
|
43 |
-
for result in features_data['results']:
|
44 |
-
if result['token'] == '<bos>':
|
45 |
-
continue
|
46 |
-
|
47 |
-
gr.Markdown(f"### {result['token']}")
|
48 |
-
for i, feature in enumerate(result['top_features'][:3]):
|
49 |
-
feature_id = feature['feature_index']
|
50 |
-
if first_feature_id is None:
|
51 |
-
first_feature_id = feature_id
|
52 |
-
gr.Button(
|
53 |
-
f"Feature {feature_id} (Activation: {feature['activation_value']:.2f})",
|
54 |
-
elem_id=str(feature_id)
|
55 |
-
).click(
|
56 |
-
fn=lambda fid=feature_id: create_dashboard(fid),
|
57 |
-
outputs=dashboard
|
58 |
-
)
|
59 |
|
60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
|
62 |
css = """
|
63 |
@import url('https://fonts.googleapis.com/css2?family=Open+Sans:wght@300;400;600;700&display=swap');
|
|
|
64 |
body { font-family: 'Open Sans', sans-serif !important; }
|
|
|
65 |
.dashboard-container {
|
66 |
border: 1px solid #e0e5ff;
|
67 |
border-radius: 8px;
|
68 |
background-color: #ffffff;
|
69 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
"""
|
71 |
|
72 |
theme = gr.themes.Soft(
|
@@ -79,6 +105,9 @@ theme = gr.themes.Soft(
|
|
79 |
)
|
80 |
)
|
81 |
|
|
|
|
|
|
|
82 |
with gr.Blocks(theme=theme, css=css) as demo:
|
83 |
gr.Markdown("# Brand Analyzer", elem_classes="text-2xl font-bold mb-2")
|
84 |
gr.Markdown("*Analyze text using Gemma's interpretable neural features*", elem_classes="text-gray-600 mb-6")
|
@@ -97,13 +126,16 @@ with gr.Blocks(theme=theme, css=css) as demo:
|
|
97 |
)
|
98 |
|
99 |
with gr.Column(scale=2):
|
100 |
-
|
101 |
dashboard = gr.HTML()
|
|
|
|
|
|
|
102 |
|
103 |
analyze_btn.click(
|
104 |
-
fn=
|
105 |
inputs=[input_text],
|
106 |
-
outputs=[
|
107 |
)
|
108 |
|
109 |
if __name__ == "__main__":
|
|
|
1 |
import gradio as gr
|
2 |
import requests
|
|
|
3 |
|
4 |
+
def get_features(text: str):
|
5 |
url = "https://www.neuronpedia.org/api/search-with-topk"
|
6 |
payload = {
|
7 |
"modelId": "gemma-2-2b",
|
|
|
29 |
</div>
|
30 |
"""
|
31 |
|
32 |
+
def analyze_text(text: str):
|
33 |
if not text:
|
34 |
+
return gr.update(visible=False), ""
|
35 |
+
|
36 |
features_data = get_features(text)
|
37 |
if not features_data:
|
38 |
+
return gr.update(visible=False), ""
|
39 |
+
|
40 |
+
html = "<div class='features-list'>"
|
41 |
first_feature_id = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
|
43 |
+
for result in features_data['results']:
|
44 |
+
if result['token'] == '<bos>':
|
45 |
+
continue
|
46 |
+
|
47 |
+
token = result['token']
|
48 |
+
html += f"<h3>{token}</h3>"
|
49 |
+
|
50 |
+
for feature in result['top_features'][:3]:
|
51 |
+
feature_id = feature['feature_index']
|
52 |
+
if first_feature_id is None:
|
53 |
+
first_feature_id = feature_id
|
54 |
+
|
55 |
+
html += f"""
|
56 |
+
<button onclick='document.dispatchEvent(new CustomEvent("select_feature",
|
57 |
+
{{detail: {{feature_id: {feature_id}}}}}))' class='feature-button'>
|
58 |
+
Feature {feature_id} (Activation: {feature['activation_value']:.2f})
|
59 |
+
</button>
|
60 |
+
"""
|
61 |
+
|
62 |
+
html += "</div>"
|
63 |
+
initial_dashboard = create_dashboard(first_feature_id) if first_feature_id else ""
|
64 |
+
|
65 |
+
return html, initial_dashboard
|
66 |
|
67 |
css = """
|
68 |
@import url('https://fonts.googleapis.com/css2?family=Open+Sans:wght@300;400;600;700&display=swap');
|
69 |
+
|
70 |
body { font-family: 'Open Sans', sans-serif !important; }
|
71 |
+
|
72 |
.dashboard-container {
|
73 |
border: 1px solid #e0e5ff;
|
74 |
border-radius: 8px;
|
75 |
background-color: #ffffff;
|
76 |
}
|
77 |
+
|
78 |
+
.features-list h3 {
|
79 |
+
margin-top: 1rem;
|
80 |
+
font-weight: 600;
|
81 |
+
}
|
82 |
+
|
83 |
+
.feature-button {
|
84 |
+
display: block;
|
85 |
+
margin: 0.5rem 0;
|
86 |
+
padding: 0.5rem 1rem;
|
87 |
+
background-color: #f3f4f6;
|
88 |
+
border: 1px solid #e5e7eb;
|
89 |
+
border-radius: 0.375rem;
|
90 |
+
cursor: pointer;
|
91 |
+
}
|
92 |
+
|
93 |
+
.feature-button:hover {
|
94 |
+
background-color: #e5e7eb;
|
95 |
+
}
|
96 |
"""
|
97 |
|
98 |
theme = gr.themes.Soft(
|
|
|
105 |
)
|
106 |
)
|
107 |
|
108 |
+
def update_dashboard(feature_id: int):
|
109 |
+
return create_dashboard(feature_id)
|
110 |
+
|
111 |
with gr.Blocks(theme=theme, css=css) as demo:
|
112 |
gr.Markdown("# Brand Analyzer", elem_classes="text-2xl font-bold mb-2")
|
113 |
gr.Markdown("*Analyze text using Gemma's interpretable neural features*", elem_classes="text-gray-600 mb-6")
|
|
|
126 |
)
|
127 |
|
128 |
with gr.Column(scale=2):
|
129 |
+
features_html = gr.HTML()
|
130 |
dashboard = gr.HTML()
|
131 |
+
|
132 |
+
# Handle feature selection via JavaScript
|
133 |
+
dashboard.change(fn=update_dashboard, inputs=gr.Textbox(visible=False), outputs=dashboard)
|
134 |
|
135 |
analyze_btn.click(
|
136 |
+
fn=analyze_text,
|
137 |
inputs=[input_text],
|
138 |
+
outputs=[features_html, dashboard]
|
139 |
)
|
140 |
|
141 |
if __name__ == "__main__":
|