Corey Morris
commited on
Commit
•
8488477
1
Parent(s):
ca8e784
Hiding filters unless box is selected. Removed model name column because it is the index of the table
Browse files
app.py
CHANGED
@@ -45,6 +45,9 @@ class MultiURLData:
|
|
45 |
cols = cols[-1:] + cols[:-1]
|
46 |
data = data[cols]
|
47 |
|
|
|
|
|
|
|
48 |
# create a new column that averages the results from each of the columns with a name that start with MMLU
|
49 |
data['MMLU_average'] = data.filter(regex='MMLU').mean(axis=1)
|
50 |
|
@@ -56,109 +59,96 @@ class MultiURLData:
|
|
56 |
|
57 |
return data
|
58 |
|
59 |
-
|
60 |
-
|
61 |
def get_data(self, selected_models):
|
62 |
-
filtered_data = self.data[self.data
|
63 |
return filtered_data
|
64 |
|
65 |
data_provider = MultiURLData()
|
66 |
|
67 |
-
st.title('
|
68 |
|
69 |
-
# TODO actually use these checkboxes as filters
|
70 |
-
## Desired behavior
|
71 |
-
## model and column selection is hidden by default
|
72 |
-
## when the user clicks the checkbox, the model and column selection appears
|
73 |
filters = st.checkbox('Add filters')
|
74 |
|
75 |
-
# Create
|
76 |
-
selected_columns =
|
77 |
-
|
78 |
-
data_provider.data.columns.tolist(),
|
79 |
-
default=data_provider.data.columns.tolist()
|
80 |
-
)
|
81 |
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
)
|
|
|
|
|
87 |
|
|
|
|
|
|
|
|
|
|
|
88 |
|
89 |
# Get the filtered data and display it in a table
|
90 |
st.header('Sortable table')
|
91 |
filtered_data = data_provider.get_data(selected_models)
|
92 |
-
st.dataframe(filtered_data)
|
93 |
|
94 |
-
|
95 |
-
|
|
|
|
|
|
|
|
|
|
|
96 |
if models is not None:
|
97 |
-
df = df[df
|
98 |
|
99 |
-
# Create a plot with new data
|
100 |
plot_data = pd.DataFrame({
|
101 |
-
'Model':
|
102 |
-
arc_column:
|
103 |
-
moral_column:
|
104 |
})
|
105 |
|
106 |
-
# Calculate color column
|
107 |
plot_data['color'] = 'purple'
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
fig = px.scatter(plot_data, x=arc_column, y=moral_column, color='color', hover_data=['Model'], trendline="ols") #other option ols
|
115 |
-
fig.update_layout(showlegend=False, # hide legend
|
116 |
-
xaxis_title=arc_column,
|
117 |
-
yaxis_title=moral_column,
|
118 |
-
xaxis = dict(),
|
119 |
-
yaxis = dict())
|
120 |
|
121 |
return fig
|
122 |
|
123 |
|
124 |
-
# models_to_plot = ['Model1', 'Model2', 'Model3']
|
125 |
-
# fig = create_plot(filtered_data, 'Model Name', 'arc:challenge|25', 'moral_scenarios|5', models=models_to_plot)
|
126 |
|
127 |
st.header('Overall benchmark comparison')
|
128 |
|
129 |
-
fig = create_plot(filtered_data, '
|
130 |
st.plotly_chart(fig)
|
131 |
|
132 |
-
fig = create_plot(filtered_data, '
|
133 |
st.plotly_chart(fig)
|
134 |
|
135 |
-
fig = create_plot(filtered_data, '
|
136 |
st.plotly_chart(fig)
|
137 |
|
138 |
-
# create a new dataframe that only has the 50 highest performing models on MMLU_average
|
139 |
st.header('Top 50 models on MMLU_average')
|
140 |
top_50 = filtered_data.nlargest(50, 'MMLU_average')
|
141 |
-
fig = create_plot(top_50, '
|
142 |
st.plotly_chart(fig)
|
143 |
|
144 |
-
# Add heading to page to say Moral Scenarios
|
145 |
st.header('Moral Scenarios')
|
146 |
|
147 |
-
fig = create_plot(filtered_data, '
|
148 |
st.plotly_chart(fig)
|
149 |
|
150 |
-
|
151 |
-
fig = create_plot(filtered_data, 'Model Name', 'MMLU_moral_disputes', 'MMLU_moral_scenarios')
|
152 |
st.plotly_chart(fig)
|
153 |
|
154 |
-
fig = create_plot(filtered_data, '
|
155 |
st.plotly_chart(fig)
|
156 |
|
157 |
-
# create a histogram of moral scenarios
|
158 |
fig = px.histogram(filtered_data, x="MMLU_moral_scenarios", marginal="rug", hover_data=filtered_data.columns)
|
159 |
st.plotly_chart(fig)
|
160 |
|
161 |
-
# create a histogram of moral disputes
|
162 |
fig = px.histogram(filtered_data, x="MMLU_moral_disputes", marginal="rug", hover_data=filtered_data.columns)
|
163 |
-
st.plotly_chart(fig)
|
164 |
-
|
|
|
45 |
cols = cols[-1:] + cols[:-1]
|
46 |
data = data[cols]
|
47 |
|
48 |
+
# remove the Model Name column
|
49 |
+
data = data.drop(['Model Name'], axis=1)
|
50 |
+
|
51 |
# create a new column that averages the results from each of the columns with a name that start with MMLU
|
52 |
data['MMLU_average'] = data.filter(regex='MMLU').mean(axis=1)
|
53 |
|
|
|
59 |
|
60 |
return data
|
61 |
|
62 |
+
# filter data based on the index
|
|
|
63 |
def get_data(self, selected_models):
|
64 |
+
filtered_data = self.data[self.data.index.isin(selected_models)]
|
65 |
return filtered_data
|
66 |
|
67 |
data_provider = MultiURLData()
|
68 |
|
69 |
+
st.title('Hugging Face Model Benchmarking including MMLU by task data')
|
70 |
|
|
|
|
|
|
|
|
|
71 |
filters = st.checkbox('Add filters')
|
72 |
|
73 |
+
# Create defaults for selected columns and models
|
74 |
+
selected_columns = data_provider.data.columns.tolist()
|
75 |
+
selected_models = data_provider.data.index.tolist()
|
|
|
|
|
|
|
76 |
|
77 |
+
if filters:
|
78 |
+
# Create checkboxes for each column
|
79 |
+
selected_columns = st.multiselect(
|
80 |
+
'Select Columns',
|
81 |
+
data_provider.data.columns.tolist(),
|
82 |
+
default=selected_columns
|
83 |
+
)
|
84 |
|
85 |
+
selected_models = st.multiselect(
|
86 |
+
'Select Models',
|
87 |
+
data_provider.data.index.tolist(),
|
88 |
+
default=selected_models
|
89 |
+
)
|
90 |
|
91 |
# Get the filtered data and display it in a table
|
92 |
st.header('Sortable table')
|
93 |
filtered_data = data_provider.get_data(selected_models)
|
|
|
94 |
|
95 |
+
# sort the table by the MMLU_average column
|
96 |
+
filtered_data = filtered_data.sort_values(by=['MMLU_average'], ascending=False)
|
97 |
+
st.dataframe(filtered_data[selected_columns])
|
98 |
+
|
99 |
+
# The rest of your plotting code...
|
100 |
+
|
101 |
+
def create_plot(df, arc_column, moral_column, models=None):
|
102 |
if models is not None:
|
103 |
+
df = df[df.index.isin(models)]
|
104 |
|
|
|
105 |
plot_data = pd.DataFrame({
|
106 |
+
'Model': df.index,
|
107 |
+
arc_column: df[arc_column],
|
108 |
+
moral_column: df[moral_column],
|
109 |
})
|
110 |
|
|
|
111 |
plot_data['color'] = 'purple'
|
112 |
+
fig = px.scatter(plot_data, x=arc_column, y=moral_column, color='color', hover_data=['Model'], trendline="ols")
|
113 |
+
fig.update_layout(showlegend=False,
|
114 |
+
xaxis_title=arc_column,
|
115 |
+
yaxis_title=moral_column,
|
116 |
+
xaxis = dict(),
|
117 |
+
yaxis = dict())
|
|
|
|
|
|
|
|
|
|
|
|
|
118 |
|
119 |
return fig
|
120 |
|
121 |
|
|
|
|
|
122 |
|
123 |
st.header('Overall benchmark comparison')
|
124 |
|
125 |
+
fig = create_plot(filtered_data, 'arc:challenge|25', 'hellaswag|10')
|
126 |
st.plotly_chart(fig)
|
127 |
|
128 |
+
fig = create_plot(filtered_data, 'arc:challenge|25', 'MMLU_average')
|
129 |
st.plotly_chart(fig)
|
130 |
|
131 |
+
fig = create_plot(filtered_data, 'hellaswag|10', 'MMLU_average')
|
132 |
st.plotly_chart(fig)
|
133 |
|
|
|
134 |
st.header('Top 50 models on MMLU_average')
|
135 |
top_50 = filtered_data.nlargest(50, 'MMLU_average')
|
136 |
+
fig = create_plot(top_50, 'arc:challenge|25', 'MMLU_average')
|
137 |
st.plotly_chart(fig)
|
138 |
|
|
|
139 |
st.header('Moral Scenarios')
|
140 |
|
141 |
+
fig = create_plot(filtered_data, 'arc:challenge|25', 'MMLU_moral_scenarios')
|
142 |
st.plotly_chart(fig)
|
143 |
|
144 |
+
fig = create_plot(filtered_data, 'MMLU_moral_disputes', 'MMLU_moral_scenarios')
|
|
|
145 |
st.plotly_chart(fig)
|
146 |
|
147 |
+
fig = create_plot(filtered_data, 'MMLU_average', 'MMLU_moral_scenarios')
|
148 |
st.plotly_chart(fig)
|
149 |
|
|
|
150 |
fig = px.histogram(filtered_data, x="MMLU_moral_scenarios", marginal="rug", hover_data=filtered_data.columns)
|
151 |
st.plotly_chart(fig)
|
152 |
|
|
|
153 |
fig = px.histogram(filtered_data, x="MMLU_moral_disputes", marginal="rug", hover_data=filtered_data.columns)
|
154 |
+
st.plotly_chart(fig)
|
|