Update app.py
Browse files
app.py
CHANGED
|
@@ -60,32 +60,24 @@ with demo:
|
|
| 60 |
gr.Markdown("## UGI Leaderboard", elem_classes="text-lg text-center")
|
| 61 |
gr.Markdown("""
|
| 62 |
UGI: Uncensored General Intelligence. The average of 5 different subjects that LLMs are commonly steered away from. The leaderboard is made from roughly 60 questions overall, measuring both "willingness to answer" and "accuracy" in fact-based controversial questions.
|
| 63 |
-
|
| 64 |
Willingness: A more narrow score, solely measuring the LLM's willingness to answer controversial questions.
|
| 65 |
-
|
| 66 |
Unruly: Knowledge of activities that are generally frowned upon.
|
| 67 |
-
|
| 68 |
Internet: Knowledge of various internet information, from professional to deviant.
|
| 69 |
-
|
| 70 |
CrimeStats: Knowledge of crime statistics which are uncomfortable to talk about.
|
| 71 |
-
|
| 72 |
Stories/Jokes: Ability to write offensive stories and jokes.
|
| 73 |
-
|
| 74 |
PolContro: Knowledge of politically/socially controversial information.
|
| 75 |
""")
|
| 76 |
with gr.Column():
|
| 77 |
with gr.Row():
|
| 78 |
search_bar = gr.Textbox(placeholder=" 🔍 Search for a model...", show_label=False)
|
| 79 |
-
with gr.Row(
|
| 80 |
-
gr.
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
param_range_7 = gr.Checkbox(label="~50", value=False)
|
| 88 |
-
param_range_8 = gr.Checkbox(label="~70+", value=False)
|
| 89 |
|
| 90 |
# Load the initial leaderboard data
|
| 91 |
leaderboard_df = load_leaderboard_data("ugi-leaderboard-data.csv")
|
|
@@ -104,48 +96,22 @@ with demo:
|
|
| 104 |
# Define the search and filter functionality
|
| 105 |
inputs = [
|
| 106 |
search_bar,
|
| 107 |
-
|
| 108 |
-
param_range_2,
|
| 109 |
-
param_range_3,
|
| 110 |
-
param_range_4,
|
| 111 |
-
param_range_5,
|
| 112 |
-
param_range_6,
|
| 113 |
-
param_range_7,
|
| 114 |
-
param_range_8
|
| 115 |
]
|
| 116 |
|
| 117 |
outputs = leaderboard_table
|
| 118 |
|
| 119 |
search_bar.change(
|
| 120 |
-
fn=lambda query,
|
| 121 |
-
'~1.5': r1,
|
| 122 |
-
'~3': r2,
|
| 123 |
-
'~7': r3,
|
| 124 |
-
'~13': r4,
|
| 125 |
-
'~20': r5,
|
| 126 |
-
'~34': r6,
|
| 127 |
-
'~50': r7,
|
| 128 |
-
'~70+': r8
|
| 129 |
-
}),
|
| 130 |
inputs=inputs,
|
| 131 |
outputs=outputs
|
| 132 |
)
|
| 133 |
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
'~7': r3,
|
| 140 |
-
'~13': r4,
|
| 141 |
-
'~20': r5,
|
| 142 |
-
'~34': r6,
|
| 143 |
-
'~50': r7,
|
| 144 |
-
'~70+': r8
|
| 145 |
-
}),
|
| 146 |
-
inputs=inputs,
|
| 147 |
-
outputs=outputs
|
| 148 |
-
)
|
| 149 |
|
| 150 |
# Launch the Gradio app
|
| 151 |
demo.launch()
|
|
|
|
| 60 |
gr.Markdown("## UGI Leaderboard", elem_classes="text-lg text-center")
|
| 61 |
gr.Markdown("""
|
| 62 |
UGI: Uncensored General Intelligence. The average of 5 different subjects that LLMs are commonly steered away from. The leaderboard is made from roughly 60 questions overall, measuring both "willingness to answer" and "accuracy" in fact-based controversial questions.
|
|
|
|
| 63 |
Willingness: A more narrow score, solely measuring the LLM's willingness to answer controversial questions.
|
|
|
|
| 64 |
Unruly: Knowledge of activities that are generally frowned upon.
|
|
|
|
| 65 |
Internet: Knowledge of various internet information, from professional to deviant.
|
|
|
|
| 66 |
CrimeStats: Knowledge of crime statistics which are uncomfortable to talk about.
|
|
|
|
| 67 |
Stories/Jokes: Ability to write offensive stories and jokes.
|
|
|
|
| 68 |
PolContro: Knowledge of politically/socially controversial information.
|
| 69 |
""")
|
| 70 |
with gr.Column():
|
| 71 |
with gr.Row():
|
| 72 |
search_bar = gr.Textbox(placeholder=" 🔍 Search for a model...", show_label=False)
|
| 73 |
+
with gr.Row():
|
| 74 |
+
filter_columns_size = gr.CheckboxGroup(
|
| 75 |
+
label="Model sizes (in billions of parameters)",
|
| 76 |
+
choices=['~1.5', '~3', '~7', '~13', '~20', '~34', '~50', '~70+'],
|
| 77 |
+
value=['~1.5', '~3', '~7', '~13', '~20', '~34', '~50', '~70+'],
|
| 78 |
+
interactive=True,
|
| 79 |
+
elem_id="filter-columns-size",
|
| 80 |
+
)
|
|
|
|
|
|
|
| 81 |
|
| 82 |
# Load the initial leaderboard data
|
| 83 |
leaderboard_df = load_leaderboard_data("ugi-leaderboard-data.csv")
|
|
|
|
| 96 |
# Define the search and filter functionality
|
| 97 |
inputs = [
|
| 98 |
search_bar,
|
| 99 |
+
filter_columns_size
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
]
|
| 101 |
|
| 102 |
outputs = leaderboard_table
|
| 103 |
|
| 104 |
search_bar.change(
|
| 105 |
+
fn=lambda query, param_ranges: update_table(leaderboard_df, query, dict(zip(['~1.5', '~3', '~7', '~13', '~20', '~34', '~50', '~70+'], param_ranges))),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
inputs=inputs,
|
| 107 |
outputs=outputs
|
| 108 |
)
|
| 109 |
|
| 110 |
+
filter_columns_size.change(
|
| 111 |
+
fn=lambda query, param_ranges: update_table(leaderboard_df, query, dict(zip(['~1.5', '~3', '~7', '~13', '~20', '~34', '~50', '~70+'], param_ranges))),
|
| 112 |
+
inputs=inputs,
|
| 113 |
+
outputs=outputs
|
| 114 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
|
| 116 |
# Launch the Gradio app
|
| 117 |
demo.launch()
|