Commit
Β·
f89f357
1
Parent(s):
bd17252
nits
Browse files
app.py
CHANGED
|
@@ -138,10 +138,24 @@ prefs_data = load_all_data(repo_dir_rewardbench, subdir="pref-sets").sort_values
|
|
| 138 |
|
| 139 |
rewardbench_data_avg = avg_over_rewardbench(rewardbench_data, prefs_data).sort_values(by='average', ascending=False)
|
| 140 |
|
| 141 |
-
|
| 142 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
cols_rewardbench_data_length = ["markdown"] + ["number"] * (len(rewardbench_data_length.columns) - 1)
|
| 144 |
-
col_types_prefs = ["markdown"] + ["number"] * (len(prefs_data.columns) - 1)
|
| 145 |
# col_types_prefs_sub = ["markdown"] + ["number"] * (len(prefs_data_sub.columns) - 1)
|
| 146 |
|
| 147 |
# for showing random samples
|
|
@@ -175,36 +189,39 @@ def regex_table(dataframe, regex, filter_button):
|
|
| 175 |
# if filter_button, remove all rows with "ai2" in the model name
|
| 176 |
if isinstance(filter_button, list) or isinstance(filter_button, str):
|
| 177 |
if "AI2 Experiments" not in filter_button and ("ai2" not in regex):
|
| 178 |
-
dataframe = dataframe[~dataframe["
|
| 179 |
if "Seq. Classifiers" not in filter_button:
|
| 180 |
-
dataframe = dataframe[~dataframe["
|
| 181 |
if "DPO" not in filter_button:
|
| 182 |
-
dataframe = dataframe[~dataframe["
|
| 183 |
if "Custom Classifiers" not in filter_button:
|
| 184 |
-
dataframe = dataframe[~dataframe["
|
| 185 |
# Filter the dataframe such that 'model' contains any of the regex patterns
|
| 186 |
-
return dataframe[dataframe["
|
| 187 |
|
| 188 |
|
| 189 |
with gr.Blocks(css=custom_css) as app:
|
| 190 |
# create tabs for the app, moving the current table to one titled "rewardbench" and the benchmark_text to a tab called "About"
|
| 191 |
with gr.Row():
|
| 192 |
-
with gr.Column(scale=
|
| 193 |
# search = gr.Textbox(label="Model Search (delimit with , )", placeholder="Regex search for a model")
|
| 194 |
# filter_button = gr.Checkbox(label="Include AI2 training runs (or type ai2 above).", interactive=True)
|
| 195 |
# img = gr.Image(value="https://private-user-images.githubusercontent.com/10695622/310698241-24ed272a-0844-451f-b414-fde57478703e.png", width=500)
|
| 196 |
gr.Markdown("""
|
| 197 |

|
| 198 |
""")
|
| 199 |
-
with gr.Column(scale=
|
| 200 |
gr.Markdown(TOP_TEXT)
|
| 201 |
with gr.Tabs(elem_classes="tab-buttons") as tabs:
|
| 202 |
with gr.TabItem("π RewardBench Leaderboard"):
|
| 203 |
with gr.Row():
|
| 204 |
-
search_1 = gr.Textbox(label="Model Search (delimit with , )",
|
|
|
|
|
|
|
| 205 |
model_types_1 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "AI2 Experiments"],
|
| 206 |
value=["Seq. Classifiers", "DPO", "Custom Classifiers"],
|
| 207 |
label="Model Types",
|
|
|
|
| 208 |
# info="Which model types to include.",
|
| 209 |
)
|
| 210 |
with gr.Row():
|
|
@@ -225,10 +242,11 @@ with gr.Blocks(css=custom_css) as app:
|
|
| 225 |
|
| 226 |
with gr.TabItem("π RewardBench - Detailed"):
|
| 227 |
with gr.Row():
|
| 228 |
-
search_2 = gr.Textbox(label="Model Search (delimit with , )", placeholder="
|
| 229 |
model_types_2 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "AI2 Experiments"],
|
| 230 |
value=["Seq. Classifiers", "DPO", "Custom Classifiers"],
|
| 231 |
label="Model Types",
|
|
|
|
| 232 |
# info="Which model types to include."
|
| 233 |
)
|
| 234 |
with gr.Row():
|
|
@@ -264,10 +282,11 @@ with gr.Blocks(css=custom_css) as app:
|
|
| 264 |
# )
|
| 265 |
with gr.TabItem("Existing Test Sets"):
|
| 266 |
with gr.Row():
|
| 267 |
-
search_3 = gr.Textbox(label="Model Search (delimit with , )", placeholder="
|
| 268 |
model_types_3 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "AI2 Experiments"],
|
| 269 |
value=["Seq. Classifiers", "DPO", "Custom Classifiers"],
|
| 270 |
-
label="Model Types",
|
|
|
|
| 271 |
# info="Which model types to include.",
|
| 272 |
)
|
| 273 |
with gr.Row():
|
|
|
|
| 138 |
|
| 139 |
rewardbench_data_avg = avg_over_rewardbench(rewardbench_data, prefs_data).sort_values(by='average', ascending=False)
|
| 140 |
|
| 141 |
+
def prep_df(df):
|
| 142 |
+
# add column to 0th entry with count (column name itself empty)
|
| 143 |
+
df.insert(0, '', range(1, 1 + len(df)))
|
| 144 |
+
|
| 145 |
+
# replace "model" with "Model" and "model_type" with "Model Type" and "average" with "Average"
|
| 146 |
+
df = df.rename(columns={"model": "Model", "model_type": "Model Type", "average": "Average"})
|
| 147 |
+
return df
|
| 148 |
+
|
| 149 |
+
# add count column to all dataframes
|
| 150 |
+
rewardbench_data = prep_df(rewardbench_data)
|
| 151 |
+
rewardbench_data_avg = prep_df(rewardbench_data_avg)
|
| 152 |
+
rewardbench_data_length = prep_df(rewardbench_data_length)
|
| 153 |
+
prefs_data = prep_df(prefs_data)
|
| 154 |
+
|
| 155 |
+
col_types_rewardbench = ["number"] + ["markdown"] + ["str"] + ["number"] * (len(rewardbench_data.columns) - 1)
|
| 156 |
+
col_types_rewardbench_avg = ["number"] + ["markdown"]+ ["str"] + ["number"] * (len(rewardbench_data_avg.columns) - 1)
|
| 157 |
cols_rewardbench_data_length = ["markdown"] + ["number"] * (len(rewardbench_data_length.columns) - 1)
|
| 158 |
+
col_types_prefs = ["number"] + ["markdown"] + ["number"] * (len(prefs_data.columns) - 1)
|
| 159 |
# col_types_prefs_sub = ["markdown"] + ["number"] * (len(prefs_data_sub.columns) - 1)
|
| 160 |
|
| 161 |
# for showing random samples
|
|
|
|
| 189 |
# if filter_button, remove all rows with "ai2" in the model name
|
| 190 |
if isinstance(filter_button, list) or isinstance(filter_button, str):
|
| 191 |
if "AI2 Experiments" not in filter_button and ("ai2" not in regex):
|
| 192 |
+
dataframe = dataframe[~dataframe["Model"].str.contains("ai2", case=False, na=False)]
|
| 193 |
if "Seq. Classifiers" not in filter_button:
|
| 194 |
+
dataframe = dataframe[~dataframe["Model Type"].str.contains("Seq. Classifier", case=False, na=False)]
|
| 195 |
if "DPO" not in filter_button:
|
| 196 |
+
dataframe = dataframe[~dataframe["Model Type"].str.contains("DPO", case=False, na=False)]
|
| 197 |
if "Custom Classifiers" not in filter_button:
|
| 198 |
+
dataframe = dataframe[~dataframe["Model Type"].str.contains("Custom Classifier", case=False, na=False)]
|
| 199 |
# Filter the dataframe such that 'model' contains any of the regex patterns
|
| 200 |
+
return dataframe[dataframe["Model"].str.contains(combined_regex, case=False, na=False)]
|
| 201 |
|
| 202 |
|
| 203 |
with gr.Blocks(css=custom_css) as app:
|
| 204 |
# create tabs for the app, moving the current table to one titled "rewardbench" and the benchmark_text to a tab called "About"
|
| 205 |
with gr.Row():
|
| 206 |
+
with gr.Column(scale=3):
|
| 207 |
# search = gr.Textbox(label="Model Search (delimit with , )", placeholder="Regex search for a model")
|
| 208 |
# filter_button = gr.Checkbox(label="Include AI2 training runs (or type ai2 above).", interactive=True)
|
| 209 |
# img = gr.Image(value="https://private-user-images.githubusercontent.com/10695622/310698241-24ed272a-0844-451f-b414-fde57478703e.png", width=500)
|
| 210 |
gr.Markdown("""
|
| 211 |

|
| 212 |
""")
|
| 213 |
+
with gr.Column(scale=6):
|
| 214 |
gr.Markdown(TOP_TEXT)
|
| 215 |
with gr.Tabs(elem_classes="tab-buttons") as tabs:
|
| 216 |
with gr.TabItem("π RewardBench Leaderboard"):
|
| 217 |
with gr.Row():
|
| 218 |
+
search_1 = gr.Textbox(label="Model Search (delimit with , )",
|
| 219 |
+
placeholder="Model Search (delimit with , )",
|
| 220 |
+
show_label=False)
|
| 221 |
model_types_1 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "AI2 Experiments"],
|
| 222 |
value=["Seq. Classifiers", "DPO", "Custom Classifiers"],
|
| 223 |
label="Model Types",
|
| 224 |
+
show_label=False,
|
| 225 |
# info="Which model types to include.",
|
| 226 |
)
|
| 227 |
with gr.Row():
|
|
|
|
| 242 |
|
| 243 |
with gr.TabItem("π RewardBench - Detailed"):
|
| 244 |
with gr.Row():
|
| 245 |
+
search_2 = gr.Textbox(label="Model Search (delimit with , )", show_label=False, placeholder="Model Search (delimit with , )")
|
| 246 |
model_types_2 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "AI2 Experiments"],
|
| 247 |
value=["Seq. Classifiers", "DPO", "Custom Classifiers"],
|
| 248 |
label="Model Types",
|
| 249 |
+
show_label=False,
|
| 250 |
# info="Which model types to include."
|
| 251 |
)
|
| 252 |
with gr.Row():
|
|
|
|
| 282 |
# )
|
| 283 |
with gr.TabItem("Existing Test Sets"):
|
| 284 |
with gr.Row():
|
| 285 |
+
search_3 = gr.Textbox(label="Model Search (delimit with , )", show_label=False, placeholder="Model Search (delimit with , )")
|
| 286 |
model_types_3 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "AI2 Experiments"],
|
| 287 |
value=["Seq. Classifiers", "DPO", "Custom Classifiers"],
|
| 288 |
+
label="Model Types",
|
| 289 |
+
show_label=False,
|
| 290 |
# info="Which model types to include.",
|
| 291 |
)
|
| 292 |
with gr.Row():
|