Merge evals
Browse files- app.py +15 -3
- debug.ipynb +458 -110
app.py
CHANGED
|
@@ -11,7 +11,7 @@ Evaluation of H4 and community models across a diverse range of benchmarks from
|
|
| 11 |
"""
|
| 12 |
|
| 13 |
|
| 14 |
-
def get_leaderboard_df():
|
| 15 |
filepaths = list(Path("eval_results").rglob("*.json"))
|
| 16 |
|
| 17 |
# Parse filepaths to get unique models
|
|
@@ -66,11 +66,17 @@ def get_leaderboard_df():
|
|
| 66 |
df = df.reset_index().rename(columns={"index": "Model"}).round(2)
|
| 67 |
# Strip off date from model name
|
| 68 |
df["Model"] = df["Model"].apply(lambda x: x.rsplit("_", 1)[0])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
return df
|
| 70 |
|
| 71 |
|
| 72 |
-
def refresh():
|
| 73 |
-
return get_leaderboard_df()
|
| 74 |
|
| 75 |
|
| 76 |
# Function to update the table based on search query
|
|
@@ -94,11 +100,17 @@ with demo:
|
|
| 94 |
gr.Markdown(DESCRIPTION, elem_classes="markdown-text")
|
| 95 |
with gr.Row():
|
| 96 |
search_bar = gr.Textbox(placeholder="Search for your model...", show_label=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
with gr.Group():
|
|
|
|
| 98 |
leaderboard_table = gr.Dataframe(value=leaderboard_df, wrap=True, height=1000)
|
| 99 |
with gr.Row():
|
| 100 |
refresh_button = gr.Button("Refresh")
|
| 101 |
|
|
|
|
| 102 |
search_bar.submit(update_table, inputs=[search_bar], outputs=[leaderboard_table])
|
| 103 |
refresh_button.click(refresh, inputs=[], outputs=[leaderboard_table])
|
| 104 |
|
|
|
|
| 11 |
"""
|
| 12 |
|
| 13 |
|
| 14 |
+
def get_leaderboard_df(merge_values: bool = False):
|
| 15 |
filepaths = list(Path("eval_results").rglob("*.json"))
|
| 16 |
|
| 17 |
# Parse filepaths to get unique models
|
|
|
|
| 66 |
df = df.reset_index().rename(columns={"index": "Model"}).round(2)
|
| 67 |
# Strip off date from model name
|
| 68 |
df["Model"] = df["Model"].apply(lambda x: x.rsplit("_", 1)[0])
|
| 69 |
+
|
| 70 |
+
if merge_values:
|
| 71 |
+
merged_df = df.drop(["Date", "Average"], axis=1).groupby("Model").max().reset_index()
|
| 72 |
+
merged_df.insert(loc=0, column="Average", value=merged_df.mean(axis=1, numeric_only=True))
|
| 73 |
+
merged_df = merged_df.sort_values(by=["Average"], ascending=False).round(2)
|
| 74 |
+
df = df[["Model", "Date"]].merge(merged_df, on="Model", how="left")
|
| 75 |
return df
|
| 76 |
|
| 77 |
|
| 78 |
+
def refresh(merge_values: bool = False):
|
| 79 |
+
return get_leaderboard_df(merge_values)
|
| 80 |
|
| 81 |
|
| 82 |
# Function to update the table based on search query
|
|
|
|
| 100 |
gr.Markdown(DESCRIPTION, elem_classes="markdown-text")
|
| 101 |
with gr.Row():
|
| 102 |
search_bar = gr.Textbox(placeholder="Search for your model...", show_label=False)
|
| 103 |
+
merge_values = gr.Checkbox(
|
| 104 |
+
label="Merge evals",
|
| 105 |
+
info="Merge evals for the same model. If there are duplicates, we display the largest one.",
|
| 106 |
+
)
|
| 107 |
with gr.Group():
|
| 108 |
+
leaderboard_df = get_leaderboard_df()
|
| 109 |
leaderboard_table = gr.Dataframe(value=leaderboard_df, wrap=True, height=1000)
|
| 110 |
with gr.Row():
|
| 111 |
refresh_button = gr.Button("Refresh")
|
| 112 |
|
| 113 |
+
merge_values.change(refresh, inputs=[merge_values], outputs=[leaderboard_table])
|
| 114 |
search_bar.submit(update_table, inputs=[search_bar], outputs=[leaderboard_table])
|
| 115 |
refresh_button.click(refresh, inputs=[], outputs=[leaderboard_table])
|
| 116 |
|
debug.ipynb
CHANGED
|
@@ -2,7 +2,7 @@
|
|
| 2 |
"cells": [
|
| 3 |
{
|
| 4 |
"cell_type": "code",
|
| 5 |
-
"execution_count":
|
| 6 |
"metadata": {},
|
| 7 |
"outputs": [],
|
| 8 |
"source": [
|
|
@@ -15,7 +15,7 @@
|
|
| 15 |
},
|
| 16 |
{
|
| 17 |
"cell_type": "code",
|
| 18 |
-
"execution_count":
|
| 19 |
"metadata": {},
|
| 20 |
"outputs": [],
|
| 21 |
"source": [
|
|
@@ -44,18 +44,34 @@
|
|
| 44 |
" data = json.load(file)\n",
|
| 45 |
" first_result_key = next(iter(data[\"results\"])) # gets the first key in 'results'\n",
|
| 46 |
" # TruthfulQA has two metrics, so we need to pick the `mc2` one that's reported on the leaderboard\n",
|
| 47 |
-
" if task == \"truthfulqa\":\n",
|
| 48 |
" value = data[\"results\"][first_result_key][\"truthfulqa_mc2\"]\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
" else:\n",
|
| 50 |
-
" first_metric_key = next(
|
|
|
|
|
|
|
| 51 |
" value = data[\"results\"][first_result_key][first_metric_key] # gets the value of the first metric\n",
|
| 52 |
" df.loc[model_revision, task] = value\n",
|
| 53 |
-
"
|
|
|
|
|
|
|
|
|
|
| 54 |
" # Drop rows where every entry is NaN\n",
|
| 55 |
" df = df.dropna(how=\"all\", axis=0, subset=[c for c in df.columns if c != \"Date\"])\n",
|
| 56 |
" df.insert(loc=1, column=\"Average\", value=df.mean(axis=1, numeric_only=True))\n",
|
|
|
|
|
|
|
| 57 |
" df = df.sort_values(by=[\"Average\"], ascending=False)\n",
|
| 58 |
-
" df = df.reset_index().rename(columns={\"index\": \"Model\"}).round(
|
| 59 |
" # Strip off date from model name\n",
|
| 60 |
" df[\"Model\"] = df[\"Model\"].apply(lambda x: x.rsplit(\"_\", 1)[0])\n",
|
| 61 |
" return df"
|
|
@@ -63,7 +79,7 @@
|
|
| 63 |
},
|
| 64 |
{
|
| 65 |
"cell_type": "code",
|
| 66 |
-
"execution_count":
|
| 67 |
"metadata": {},
|
| 68 |
"outputs": [],
|
| 69 |
"source": [
|
|
@@ -72,7 +88,7 @@
|
|
| 72 |
},
|
| 73 |
{
|
| 74 |
"cell_type": "code",
|
| 75 |
-
"execution_count":
|
| 76 |
"metadata": {},
|
| 77 |
"outputs": [
|
| 78 |
{
|
|
@@ -111,68 +127,68 @@
|
|
| 111 |
" <tbody>\n",
|
| 112 |
" <tr>\n",
|
| 113 |
" <th>0</th>\n",
|
| 114 |
-
" <td>NousResearch_Nous-Hermes-2-
|
| 115 |
-
" <td>2024-03-
|
| 116 |
-
" <td>
|
| 117 |
-
" <td>
|
| 118 |
-
" <td>
|
| 119 |
-
" <td>
|
| 120 |
-
" <td>
|
| 121 |
-
" <td>
|
| 122 |
-
" <td>
|
| 123 |
-
" <td>
|
| 124 |
" </tr>\n",
|
| 125 |
" <tr>\n",
|
| 126 |
" <th>1</th>\n",
|
| 127 |
-
" <td>
|
| 128 |
-
" <td>2024-03-
|
| 129 |
-
" <td>
|
|
|
|
|
|
|
| 130 |
" <td>NaN</td>\n",
|
| 131 |
-
" <td>
|
| 132 |
-
" <td>
|
|
|
|
| 133 |
" <td>NaN</td>\n",
|
| 134 |
-
" <td>0.48</td>\n",
|
| 135 |
-
" <td>0.640</td>\n",
|
| 136 |
-
" <td>0.654</td>\n",
|
| 137 |
" </tr>\n",
|
| 138 |
" <tr>\n",
|
| 139 |
" <th>2</th>\n",
|
| 140 |
-
" <td>
|
| 141 |
" <td>2024-03-02</td>\n",
|
| 142 |
-
" <td>
|
| 143 |
-
" <td>
|
| 144 |
-
" <td>
|
| 145 |
-
" <td>
|
| 146 |
-
" <td>
|
| 147 |
-
" <td>
|
| 148 |
-
" <td>
|
| 149 |
-
" <td>
|
| 150 |
" </tr>\n",
|
| 151 |
" <tr>\n",
|
| 152 |
" <th>3</th>\n",
|
| 153 |
-
" <td>
|
| 154 |
-
" <td>2024-03-
|
| 155 |
-
" <td>
|
| 156 |
-
" <td>
|
| 157 |
-
" <td>
|
| 158 |
-
" <td>
|
| 159 |
-
" <td>
|
| 160 |
-
" <td>
|
| 161 |
-
" <td>
|
| 162 |
-
" <td>
|
| 163 |
" </tr>\n",
|
| 164 |
" <tr>\n",
|
| 165 |
" <th>4</th>\n",
|
| 166 |
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
|
| 167 |
-
" <td>2024-03-
|
| 168 |
-
" <td>
|
| 169 |
-
" <td>
|
|
|
|
|
|
|
| 170 |
" <td>NaN</td>\n",
|
| 171 |
" <td>NaN</td>\n",
|
| 172 |
-
" <td>0.761</td>\n",
|
| 173 |
-
" <td>0.42</td>\n",
|
| 174 |
-
" <td>0.654</td>\n",
|
| 175 |
" <td>NaN</td>\n",
|
|
|
|
| 176 |
" </tr>\n",
|
| 177 |
" <tr>\n",
|
| 178 |
" <th>...</th>\n",
|
|
@@ -191,11 +207,11 @@
|
|
| 191 |
" <th>269</th>\n",
|
| 192 |
" <td>HuggingFaceH4_starcoder2-15b-ift_v18.0</td>\n",
|
| 193 |
" <td>2024-03-10</td>\n",
|
| 194 |
-
" <td>
|
| 195 |
-
" <td>
|
| 196 |
" <td>NaN</td>\n",
|
| 197 |
" <td>NaN</td>\n",
|
| 198 |
-
" <td>0.
|
| 199 |
" <td>NaN</td>\n",
|
| 200 |
" <td>NaN</td>\n",
|
| 201 |
" <td>NaN</td>\n",
|
|
@@ -204,11 +220,11 @@
|
|
| 204 |
" <th>270</th>\n",
|
| 205 |
" <td>HuggingFaceH4_mistral-7b-ift_v49.0</td>\n",
|
| 206 |
" <td>2024-03-07</td>\n",
|
| 207 |
-
" <td>
|
| 208 |
-
" <td>
|
| 209 |
" <td>NaN</td>\n",
|
| 210 |
" <td>NaN</td>\n",
|
| 211 |
-
" <td>0.
|
| 212 |
" <td>NaN</td>\n",
|
| 213 |
" <td>NaN</td>\n",
|
| 214 |
" <td>NaN</td>\n",
|
|
@@ -217,8 +233,8 @@
|
|
| 217 |
" <th>271</th>\n",
|
| 218 |
" <td>HuggingFaceH4_starchat-beta_main</td>\n",
|
| 219 |
" <td>2024-03-12</td>\n",
|
| 220 |
-
" <td>
|
| 221 |
-
" <td>
|
| 222 |
" <td>NaN</td>\n",
|
| 223 |
" <td>NaN</td>\n",
|
| 224 |
" <td>NaN</td>\n",
|
|
@@ -230,11 +246,11 @@
|
|
| 230 |
" <th>272</th>\n",
|
| 231 |
" <td>HuggingFaceH4_starcoder2-15b-ift_v7.0</td>\n",
|
| 232 |
" <td>2024-03-10</td>\n",
|
| 233 |
-
" <td>
|
| 234 |
-
" <td>
|
| 235 |
" <td>NaN</td>\n",
|
| 236 |
" <td>NaN</td>\n",
|
| 237 |
-
" <td>
|
| 238 |
" <td>NaN</td>\n",
|
| 239 |
" <td>NaN</td>\n",
|
| 240 |
" <td>NaN</td>\n",
|
|
@@ -243,11 +259,11 @@
|
|
| 243 |
" <th>273</th>\n",
|
| 244 |
" <td>HuggingFaceH4_zephyr-7b-beta-ift_v1.1</td>\n",
|
| 245 |
" <td>2024-03-13</td>\n",
|
| 246 |
-
" <td>
|
| 247 |
-
" <td>
|
| 248 |
" <td>NaN</td>\n",
|
| 249 |
" <td>NaN</td>\n",
|
| 250 |
-
" <td>0.
|
| 251 |
" <td>NaN</td>\n",
|
| 252 |
" <td>NaN</td>\n",
|
| 253 |
" <td>NaN</td>\n",
|
|
@@ -259,35 +275,35 @@
|
|
| 259 |
],
|
| 260 |
"text/plain": [
|
| 261 |
" Model Date Average \\\n",
|
| 262 |
-
"0
|
| 263 |
-
"1
|
| 264 |
-
"2
|
| 265 |
-
"3
|
| 266 |
-
"4 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-
|
| 267 |
".. ... ... ... \n",
|
| 268 |
-
"269 HuggingFaceH4_starcoder2-15b-ift_v18.0 2024-03-10
|
| 269 |
-
"270 HuggingFaceH4_mistral-7b-ift_v49.0 2024-03-07
|
| 270 |
-
"271 HuggingFaceH4_starchat-beta_main 2024-03-12
|
| 271 |
-
"272 HuggingFaceH4_starcoder2-15b-ift_v7.0 2024-03-10
|
| 272 |
-
"273 HuggingFaceH4_zephyr-7b-beta-ift_v1.1 2024-03-13
|
| 273 |
"\n",
|
| 274 |
-
" Ifeval Truthfulqa Winogrande Gsm8k
|
| 275 |
-
"0
|
| 276 |
-
"1
|
| 277 |
-
"2
|
| 278 |
-
"3
|
| 279 |
-
"4
|
| 280 |
-
".. ... ... ... ...
|
| 281 |
-
"269
|
| 282 |
-
"270
|
| 283 |
-
"271
|
| 284 |
-
"272
|
| 285 |
-
"273
|
| 286 |
"\n",
|
| 287 |
"[274 rows x 10 columns]"
|
| 288 |
]
|
| 289 |
},
|
| 290 |
-
"execution_count":
|
| 291 |
"metadata": {},
|
| 292 |
"output_type": "execute_result"
|
| 293 |
}
|
|
@@ -298,7 +314,7 @@
|
|
| 298 |
},
|
| 299 |
{
|
| 300 |
"cell_type": "code",
|
| 301 |
-
"execution_count":
|
| 302 |
"metadata": {},
|
| 303 |
"outputs": [
|
| 304 |
{
|
|
@@ -323,7 +339,6 @@
|
|
| 323 |
" <tr style=\"text-align: right;\">\n",
|
| 324 |
" <th></th>\n",
|
| 325 |
" <th>Model</th>\n",
|
| 326 |
-
" <th>Average</th>\n",
|
| 327 |
" <th>Ifeval</th>\n",
|
| 328 |
" <th>Truthfulqa</th>\n",
|
| 329 |
" <th>Winogrande</th>\n",
|
|
@@ -335,50 +350,383 @@
|
|
| 335 |
" </thead>\n",
|
| 336 |
" <tbody>\n",
|
| 337 |
" <tr>\n",
|
| 338 |
-
" <th>
|
| 339 |
-
" <td>HuggingFaceH4_mistral-7b-
|
| 340 |
-
" <td>
|
| 341 |
-
" <td>
|
| 342 |
-
" <td>
|
| 343 |
-
" <td>
|
| 344 |
-
" <td>
|
| 345 |
-
" <td>
|
| 346 |
-
" <td>
|
| 347 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 348 |
" </tr>\n",
|
| 349 |
" <tr>\n",
|
| 350 |
-
" <th
|
| 351 |
-
" <td
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 352 |
" <td>NaN</td>\n",
|
| 353 |
" <td>NaN</td>\n",
|
| 354 |
" <td>NaN</td>\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 355 |
" <td>NaN</td>\n",
|
| 356 |
" <td>NaN</td>\n",
|
|
|
|
| 357 |
" <td>NaN</td>\n",
|
| 358 |
" <td>NaN</td>\n",
|
| 359 |
" <td>NaN</td>\n",
|
| 360 |
" </tr>\n",
|
| 361 |
" </tbody>\n",
|
| 362 |
"</table>\n",
|
|
|
|
| 363 |
"</div>"
|
| 364 |
],
|
| 365 |
"text/plain": [
|
| 366 |
-
"
|
| 367 |
-
"
|
| 368 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 369 |
"\n",
|
| 370 |
-
"
|
| 371 |
-
"50 0.359 0.672 0.453 0.33 0.656 0.545 \n",
|
| 372 |
-
"532 NaN NaN NaN NaN NaN NaN "
|
| 373 |
]
|
| 374 |
},
|
| 375 |
-
"execution_count":
|
| 376 |
"metadata": {},
|
| 377 |
"output_type": "execute_result"
|
| 378 |
}
|
| 379 |
],
|
| 380 |
"source": [
|
| 381 |
-
"df[
|
| 382 |
]
|
| 383 |
},
|
| 384 |
{
|
|
|
|
| 2 |
"cells": [
|
| 3 |
{
|
| 4 |
"cell_type": "code",
|
| 5 |
+
"execution_count": 2,
|
| 6 |
"metadata": {},
|
| 7 |
"outputs": [],
|
| 8 |
"source": [
|
|
|
|
| 15 |
},
|
| 16 |
{
|
| 17 |
"cell_type": "code",
|
| 18 |
+
"execution_count": 3,
|
| 19 |
"metadata": {},
|
| 20 |
"outputs": [],
|
| 21 |
"source": [
|
|
|
|
| 44 |
" data = json.load(file)\n",
|
| 45 |
" first_result_key = next(iter(data[\"results\"])) # gets the first key in 'results'\n",
|
| 46 |
" # TruthfulQA has two metrics, so we need to pick the `mc2` one that's reported on the leaderboard\n",
|
| 47 |
+
" if task.lower() == \"truthfulqa\":\n",
|
| 48 |
" value = data[\"results\"][first_result_key][\"truthfulqa_mc2\"]\n",
|
| 49 |
+
" # IFEval has several metrics but we report just the prompt-loose-acc one\n",
|
| 50 |
+
" elif task.lower() == \"ifeval\":\n",
|
| 51 |
+
" value = data[\"results\"][first_result_key][\"prompt_level_loose_acc\"]\n",
|
| 52 |
+
" # MMLU has several metrics but we report just the average one\n",
|
| 53 |
+
" elif task.lower() == \"mmlu\":\n",
|
| 54 |
+
" value = data[\"results\"][\"lighteval|mmlu:_average|5\"][\"acc\"]\n",
|
| 55 |
+
" # HellaSwag and ARC reports acc_norm\n",
|
| 56 |
+
" elif task.lower() in [\"hellaswag\", \"arc\"]:\n",
|
| 57 |
+
" value = data[\"results\"][first_result_key][\"acc_norm\"]\n",
|
| 58 |
" else:\n",
|
| 59 |
+
" first_metric_key = next(\n",
|
| 60 |
+
" iter(data[\"results\"][first_result_key])\n",
|
| 61 |
+
" ) # gets the first key in the first result\n",
|
| 62 |
" value = data[\"results\"][first_result_key][first_metric_key] # gets the value of the first metric\n",
|
| 63 |
" df.loc[model_revision, task] = value\n",
|
| 64 |
+
"\n",
|
| 65 |
+
" # Put IFEval in first column\n",
|
| 66 |
+
" ifeval_col = df.pop(\"Ifeval\")\n",
|
| 67 |
+
" df.insert(1, \"Ifeval\", ifeval_col)\n",
|
| 68 |
" # Drop rows where every entry is NaN\n",
|
| 69 |
" df = df.dropna(how=\"all\", axis=0, subset=[c for c in df.columns if c != \"Date\"])\n",
|
| 70 |
" df.insert(loc=1, column=\"Average\", value=df.mean(axis=1, numeric_only=True))\n",
|
| 71 |
+
" # Convert all values to percentage\n",
|
| 72 |
+
" df[df.select_dtypes(include=[\"number\"]).columns] *= 100.0\n",
|
| 73 |
" df = df.sort_values(by=[\"Average\"], ascending=False)\n",
|
| 74 |
+
" df = df.reset_index().rename(columns={\"index\": \"Model\"}).round(2)\n",
|
| 75 |
" # Strip off date from model name\n",
|
| 76 |
" df[\"Model\"] = df[\"Model\"].apply(lambda x: x.rsplit(\"_\", 1)[0])\n",
|
| 77 |
" return df"
|
|
|
|
| 79 |
},
|
| 80 |
{
|
| 81 |
"cell_type": "code",
|
| 82 |
+
"execution_count": 4,
|
| 83 |
"metadata": {},
|
| 84 |
"outputs": [],
|
| 85 |
"source": [
|
|
|
|
| 88 |
},
|
| 89 |
{
|
| 90 |
"cell_type": "code",
|
| 91 |
+
"execution_count": 5,
|
| 92 |
"metadata": {},
|
| 93 |
"outputs": [
|
| 94 |
{
|
|
|
|
| 127 |
" <tbody>\n",
|
| 128 |
" <tr>\n",
|
| 129 |
" <th>0</th>\n",
|
| 130 |
+
" <td>NousResearch_Nous-Hermes-2-Yi-34B_main</td>\n",
|
| 131 |
+
" <td>2024-03-04</td>\n",
|
| 132 |
+
" <td>74.01</td>\n",
|
| 133 |
+
" <td>NaN</td>\n",
|
| 134 |
+
" <td>61.44</td>\n",
|
| 135 |
+
" <td>80.58</td>\n",
|
| 136 |
+
" <td>NaN</td>\n",
|
| 137 |
+
" <td>76.24</td>\n",
|
| 138 |
+
" <td>83.79</td>\n",
|
| 139 |
+
" <td>68.00</td>\n",
|
| 140 |
" </tr>\n",
|
| 141 |
" <tr>\n",
|
| 142 |
" <th>1</th>\n",
|
| 143 |
+
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
|
| 144 |
+
" <td>2024-03-05</td>\n",
|
| 145 |
+
" <td>71.62</td>\n",
|
| 146 |
+
" <td>55.27</td>\n",
|
| 147 |
+
" <td>NaN</td>\n",
|
| 148 |
" <td>NaN</td>\n",
|
| 149 |
+
" <td>76.12</td>\n",
|
| 150 |
+
" <td>71.18</td>\n",
|
| 151 |
+
" <td>83.94</td>\n",
|
| 152 |
" <td>NaN</td>\n",
|
|
|
|
|
|
|
|
|
|
| 153 |
" </tr>\n",
|
| 154 |
" <tr>\n",
|
| 155 |
" <th>2</th>\n",
|
| 156 |
+
" <td>NousResearch_Nous-Hermes-2-Mixtral-8x7B-DPO_main</td>\n",
|
| 157 |
" <td>2024-03-02</td>\n",
|
| 158 |
+
" <td>70.43</td>\n",
|
| 159 |
+
" <td>59.33</td>\n",
|
| 160 |
+
" <td>64.76</td>\n",
|
| 161 |
+
" <td>78.53</td>\n",
|
| 162 |
+
" <td>62.17</td>\n",
|
| 163 |
+
" <td>71.96</td>\n",
|
| 164 |
+
" <td>85.42</td>\n",
|
| 165 |
+
" <td>70.82</td>\n",
|
| 166 |
" </tr>\n",
|
| 167 |
" <tr>\n",
|
| 168 |
" <th>3</th>\n",
|
| 169 |
+
" <td>mistralai_Mixtral-8x7B-Instruct-v0.1_main</td>\n",
|
| 170 |
+
" <td>2024-03-02</td>\n",
|
| 171 |
+
" <td>69.80</td>\n",
|
| 172 |
+
" <td>55.08</td>\n",
|
| 173 |
+
" <td>70.79</td>\n",
|
| 174 |
+
" <td>73.56</td>\n",
|
| 175 |
+
" <td>59.89</td>\n",
|
| 176 |
+
" <td>70.60</td>\n",
|
| 177 |
+
" <td>86.68</td>\n",
|
| 178 |
+
" <td>72.01</td>\n",
|
| 179 |
" </tr>\n",
|
| 180 |
" <tr>\n",
|
| 181 |
" <th>4</th>\n",
|
| 182 |
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
|
| 183 |
+
" <td>2024-03-04</td>\n",
|
| 184 |
+
" <td>67.03</td>\n",
|
| 185 |
+
" <td>NaN</td>\n",
|
| 186 |
+
" <td>57.78</td>\n",
|
| 187 |
+
" <td>79.16</td>\n",
|
| 188 |
" <td>NaN</td>\n",
|
| 189 |
" <td>NaN</td>\n",
|
|
|
|
|
|
|
|
|
|
| 190 |
" <td>NaN</td>\n",
|
| 191 |
+
" <td>64.16</td>\n",
|
| 192 |
" </tr>\n",
|
| 193 |
" <tr>\n",
|
| 194 |
" <th>...</th>\n",
|
|
|
|
| 207 |
" <th>269</th>\n",
|
| 208 |
" <td>HuggingFaceH4_starcoder2-15b-ift_v18.0</td>\n",
|
| 209 |
" <td>2024-03-10</td>\n",
|
| 210 |
+
" <td>11.23</td>\n",
|
| 211 |
+
" <td>21.63</td>\n",
|
| 212 |
" <td>NaN</td>\n",
|
| 213 |
" <td>NaN</td>\n",
|
| 214 |
+
" <td>0.83</td>\n",
|
| 215 |
" <td>NaN</td>\n",
|
| 216 |
" <td>NaN</td>\n",
|
| 217 |
" <td>NaN</td>\n",
|
|
|
|
| 220 |
" <th>270</th>\n",
|
| 221 |
" <td>HuggingFaceH4_mistral-7b-ift_v49.0</td>\n",
|
| 222 |
" <td>2024-03-07</td>\n",
|
| 223 |
+
" <td>10.07</td>\n",
|
| 224 |
+
" <td>20.15</td>\n",
|
| 225 |
" <td>NaN</td>\n",
|
| 226 |
" <td>NaN</td>\n",
|
| 227 |
+
" <td>0.00</td>\n",
|
| 228 |
" <td>NaN</td>\n",
|
| 229 |
" <td>NaN</td>\n",
|
| 230 |
" <td>NaN</td>\n",
|
|
|
|
| 233 |
" <th>271</th>\n",
|
| 234 |
" <td>HuggingFaceH4_starchat-beta_main</td>\n",
|
| 235 |
" <td>2024-03-12</td>\n",
|
| 236 |
+
" <td>8.13</td>\n",
|
| 237 |
+
" <td>8.13</td>\n",
|
| 238 |
" <td>NaN</td>\n",
|
| 239 |
" <td>NaN</td>\n",
|
| 240 |
" <td>NaN</td>\n",
|
|
|
|
| 246 |
" <th>272</th>\n",
|
| 247 |
" <td>HuggingFaceH4_starcoder2-15b-ift_v7.0</td>\n",
|
| 248 |
" <td>2024-03-10</td>\n",
|
| 249 |
+
" <td>7.88</td>\n",
|
| 250 |
+
" <td>12.57</td>\n",
|
| 251 |
" <td>NaN</td>\n",
|
| 252 |
" <td>NaN</td>\n",
|
| 253 |
+
" <td>3.18</td>\n",
|
| 254 |
" <td>NaN</td>\n",
|
| 255 |
" <td>NaN</td>\n",
|
| 256 |
" <td>NaN</td>\n",
|
|
|
|
| 259 |
" <th>273</th>\n",
|
| 260 |
" <td>HuggingFaceH4_zephyr-7b-beta-ift_v1.1</td>\n",
|
| 261 |
" <td>2024-03-13</td>\n",
|
| 262 |
+
" <td>4.71</td>\n",
|
| 263 |
+
" <td>9.43</td>\n",
|
| 264 |
" <td>NaN</td>\n",
|
| 265 |
" <td>NaN</td>\n",
|
| 266 |
+
" <td>0.00</td>\n",
|
| 267 |
" <td>NaN</td>\n",
|
| 268 |
" <td>NaN</td>\n",
|
| 269 |
" <td>NaN</td>\n",
|
|
|
|
| 275 |
],
|
| 276 |
"text/plain": [
|
| 277 |
" Model Date Average \\\n",
|
| 278 |
+
"0 NousResearch_Nous-Hermes-2-Yi-34B_main 2024-03-04 74.01 \n",
|
| 279 |
+
"1 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-05 71.62 \n",
|
| 280 |
+
"2 NousResearch_Nous-Hermes-2-Mixtral-8x7B-DPO_main 2024-03-02 70.43 \n",
|
| 281 |
+
"3 mistralai_Mixtral-8x7B-Instruct-v0.1_main 2024-03-02 69.80 \n",
|
| 282 |
+
"4 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-04 67.03 \n",
|
| 283 |
".. ... ... ... \n",
|
| 284 |
+
"269 HuggingFaceH4_starcoder2-15b-ift_v18.0 2024-03-10 11.23 \n",
|
| 285 |
+
"270 HuggingFaceH4_mistral-7b-ift_v49.0 2024-03-07 10.07 \n",
|
| 286 |
+
"271 HuggingFaceH4_starchat-beta_main 2024-03-12 8.13 \n",
|
| 287 |
+
"272 HuggingFaceH4_starcoder2-15b-ift_v7.0 2024-03-10 7.88 \n",
|
| 288 |
+
"273 HuggingFaceH4_zephyr-7b-beta-ift_v1.1 2024-03-13 4.71 \n",
|
| 289 |
"\n",
|
| 290 |
+
" Ifeval Truthfulqa Winogrande Gsm8k Mmlu Hellaswag Arc \n",
|
| 291 |
+
"0 NaN 61.44 80.58 NaN 76.24 83.79 68.00 \n",
|
| 292 |
+
"1 55.27 NaN NaN 76.12 71.18 83.94 NaN \n",
|
| 293 |
+
"2 59.33 64.76 78.53 62.17 71.96 85.42 70.82 \n",
|
| 294 |
+
"3 55.08 70.79 73.56 59.89 70.60 86.68 72.01 \n",
|
| 295 |
+
"4 NaN 57.78 79.16 NaN NaN NaN 64.16 \n",
|
| 296 |
+
".. ... ... ... ... ... ... ... \n",
|
| 297 |
+
"269 21.63 NaN NaN 0.83 NaN NaN NaN \n",
|
| 298 |
+
"270 20.15 NaN NaN 0.00 NaN NaN NaN \n",
|
| 299 |
+
"271 8.13 NaN NaN NaN NaN NaN NaN \n",
|
| 300 |
+
"272 12.57 NaN NaN 3.18 NaN NaN NaN \n",
|
| 301 |
+
"273 9.43 NaN NaN 0.00 NaN NaN NaN \n",
|
| 302 |
"\n",
|
| 303 |
"[274 rows x 10 columns]"
|
| 304 |
]
|
| 305 |
},
|
| 306 |
+
"execution_count": 5,
|
| 307 |
"metadata": {},
|
| 308 |
"output_type": "execute_result"
|
| 309 |
}
|
|
|
|
| 314 |
},
|
| 315 |
{
|
| 316 |
"cell_type": "code",
|
| 317 |
+
"execution_count": 14,
|
| 318 |
"metadata": {},
|
| 319 |
"outputs": [
|
| 320 |
{
|
|
|
|
| 339 |
" <tr style=\"text-align: right;\">\n",
|
| 340 |
" <th></th>\n",
|
| 341 |
" <th>Model</th>\n",
|
|
|
|
| 342 |
" <th>Ifeval</th>\n",
|
| 343 |
" <th>Truthfulqa</th>\n",
|
| 344 |
" <th>Winogrande</th>\n",
|
|
|
|
| 350 |
" </thead>\n",
|
| 351 |
" <tbody>\n",
|
| 352 |
" <tr>\n",
|
| 353 |
+
" <th>0</th>\n",
|
| 354 |
+
" <td>HuggingFaceH4_mistral-7b-ift_v41.0</td>\n",
|
| 355 |
+
" <td>44.36</td>\n",
|
| 356 |
+
" <td>49.35</td>\n",
|
| 357 |
+
" <td>72.93</td>\n",
|
| 358 |
+
" <td>37.30</td>\n",
|
| 359 |
+
" <td>60.82</td>\n",
|
| 360 |
+
" <td>79.70</td>\n",
|
| 361 |
+
" <td>58.36</td>\n",
|
| 362 |
+
" </tr>\n",
|
| 363 |
+
" <tr>\n",
|
| 364 |
+
" <th>1</th>\n",
|
| 365 |
+
" <td>HuggingFaceH4_mistral-7b-ift_v41.1</td>\n",
|
| 366 |
+
" <td>47.32</td>\n",
|
| 367 |
+
" <td>47.89</td>\n",
|
| 368 |
+
" <td>72.69</td>\n",
|
| 369 |
+
" <td>36.32</td>\n",
|
| 370 |
+
" <td>60.34</td>\n",
|
| 371 |
+
" <td>79.57</td>\n",
|
| 372 |
+
" <td>57.51</td>\n",
|
| 373 |
+
" </tr>\n",
|
| 374 |
+
" <tr>\n",
|
| 375 |
+
" <th>2</th>\n",
|
| 376 |
+
" <td>HuggingFaceH4_mistral-7b-ift_v41.10</td>\n",
|
| 377 |
+
" <td>32.72</td>\n",
|
| 378 |
+
" <td>51.05</td>\n",
|
| 379 |
+
" <td>72.45</td>\n",
|
| 380 |
+
" <td>25.93</td>\n",
|
| 381 |
+
" <td>59.75</td>\n",
|
| 382 |
+
" <td>81.92</td>\n",
|
| 383 |
+
" <td>59.22</td>\n",
|
| 384 |
+
" </tr>\n",
|
| 385 |
+
" <tr>\n",
|
| 386 |
+
" <th>3</th>\n",
|
| 387 |
+
" <td>HuggingFaceH4_mistral-7b-ift_v41.11</td>\n",
|
| 388 |
+
" <td>37.89</td>\n",
|
| 389 |
+
" <td>51.05</td>\n",
|
| 390 |
+
" <td>64.56</td>\n",
|
| 391 |
+
" <td>17.59</td>\n",
|
| 392 |
+
" <td>57.60</td>\n",
|
| 393 |
+
" <td>77.65</td>\n",
|
| 394 |
+
" <td>55.89</td>\n",
|
| 395 |
+
" </tr>\n",
|
| 396 |
+
" <tr>\n",
|
| 397 |
+
" <th>4</th>\n",
|
| 398 |
+
" <td>HuggingFaceH4_mistral-7b-ift_v41.12</td>\n",
|
| 399 |
+
" <td>37.89</td>\n",
|
| 400 |
+
" <td>45.94</td>\n",
|
| 401 |
+
" <td>63.30</td>\n",
|
| 402 |
+
" <td>21.15</td>\n",
|
| 403 |
+
" <td>58.50</td>\n",
|
| 404 |
+
" <td>74.94</td>\n",
|
| 405 |
+
" <td>52.73</td>\n",
|
| 406 |
" </tr>\n",
|
| 407 |
" <tr>\n",
|
| 408 |
+
" <th>...</th>\n",
|
| 409 |
+
" <td>...</td>\n",
|
| 410 |
+
" <td>...</td>\n",
|
| 411 |
+
" <td>...</td>\n",
|
| 412 |
+
" <td>...</td>\n",
|
| 413 |
+
" <td>...</td>\n",
|
| 414 |
+
" <td>...</td>\n",
|
| 415 |
+
" <td>...</td>\n",
|
| 416 |
+
" <td>...</td>\n",
|
| 417 |
+
" </tr>\n",
|
| 418 |
+
" <tr>\n",
|
| 419 |
+
" <th>258</th>\n",
|
| 420 |
+
" <td>mistralai_Mistral-7B-Instruct-v0.2_main</td>\n",
|
| 421 |
+
" <td>53.97</td>\n",
|
| 422 |
+
" <td>70.68</td>\n",
|
| 423 |
+
" <td>68.82</td>\n",
|
| 424 |
+
" <td>38.13</td>\n",
|
| 425 |
+
" <td>59.43</td>\n",
|
| 426 |
+
" <td>83.45</td>\n",
|
| 427 |
+
" <td>65.70</td>\n",
|
| 428 |
+
" </tr>\n",
|
| 429 |
+
" <tr>\n",
|
| 430 |
+
" <th>259</th>\n",
|
| 431 |
+
" <td>mistralai_Mixtral-8x7B-Instruct-v0.1_main</td>\n",
|
| 432 |
+
" <td>55.08</td>\n",
|
| 433 |
+
" <td>70.79</td>\n",
|
| 434 |
+
" <td>73.56</td>\n",
|
| 435 |
+
" <td>59.89</td>\n",
|
| 436 |
+
" <td>70.60</td>\n",
|
| 437 |
+
" <td>86.68</td>\n",
|
| 438 |
+
" <td>72.01</td>\n",
|
| 439 |
+
" </tr>\n",
|
| 440 |
+
" <tr>\n",
|
| 441 |
+
" <th>260</th>\n",
|
| 442 |
+
" <td>openchat_openchat-3.5-0106_main</td>\n",
|
| 443 |
+
" <td>54.71</td>\n",
|
| 444 |
+
" <td>57.55</td>\n",
|
| 445 |
+
" <td>72.53</td>\n",
|
| 446 |
+
" <td>66.19</td>\n",
|
| 447 |
+
" <td>63.72</td>\n",
|
| 448 |
+
" <td>80.10</td>\n",
|
| 449 |
+
" <td>61.01</td>\n",
|
| 450 |
+
" </tr>\n",
|
| 451 |
+
" <tr>\n",
|
| 452 |
+
" <th>261</th>\n",
|
| 453 |
+
" <td>stabilityai_stablelm-zephyr-3b_main</td>\n",
|
| 454 |
+
" <td>34.75</td>\n",
|
| 455 |
+
" <td>46.19</td>\n",
|
| 456 |
+
" <td>58.41</td>\n",
|
| 457 |
+
" <td>40.18</td>\n",
|
| 458 |
+
" <td>45.18</td>\n",
|
| 459 |
+
" <td>71.57</td>\n",
|
| 460 |
+
" <td>45.82</td>\n",
|
| 461 |
+
" </tr>\n",
|
| 462 |
+
" <tr>\n",
|
| 463 |
+
" <th>262</th>\n",
|
| 464 |
+
" <td>teknium_OpenHermes-2.5-Mistral-7B_main</td>\n",
|
| 465 |
+
" <td>52.68</td>\n",
|
| 466 |
+
" <td>58.62</td>\n",
|
| 467 |
+
" <td>72.14</td>\n",
|
| 468 |
+
" <td>54.06</td>\n",
|
| 469 |
+
" <td>63.01</td>\n",
|
| 470 |
+
" <td>82.34</td>\n",
|
| 471 |
+
" <td>62.97</td>\n",
|
| 472 |
+
" </tr>\n",
|
| 473 |
+
" </tbody>\n",
|
| 474 |
+
"</table>\n",
|
| 475 |
+
"<p>263 rows × 8 columns</p>\n",
|
| 476 |
+
"</div>"
|
| 477 |
+
],
|
| 478 |
+
"text/plain": [
|
| 479 |
+
" Model Ifeval Truthfulqa \\\n",
|
| 480 |
+
"0 HuggingFaceH4_mistral-7b-ift_v41.0 44.36 49.35 \n",
|
| 481 |
+
"1 HuggingFaceH4_mistral-7b-ift_v41.1 47.32 47.89 \n",
|
| 482 |
+
"2 HuggingFaceH4_mistral-7b-ift_v41.10 32.72 51.05 \n",
|
| 483 |
+
"3 HuggingFaceH4_mistral-7b-ift_v41.11 37.89 51.05 \n",
|
| 484 |
+
"4 HuggingFaceH4_mistral-7b-ift_v41.12 37.89 45.94 \n",
|
| 485 |
+
".. ... ... ... \n",
|
| 486 |
+
"258 mistralai_Mistral-7B-Instruct-v0.2_main 53.97 70.68 \n",
|
| 487 |
+
"259 mistralai_Mixtral-8x7B-Instruct-v0.1_main 55.08 70.79 \n",
|
| 488 |
+
"260 openchat_openchat-3.5-0106_main 54.71 57.55 \n",
|
| 489 |
+
"261 stabilityai_stablelm-zephyr-3b_main 34.75 46.19 \n",
|
| 490 |
+
"262 teknium_OpenHermes-2.5-Mistral-7B_main 52.68 58.62 \n",
|
| 491 |
+
"\n",
|
| 492 |
+
" Winogrande Gsm8k Mmlu Hellaswag Arc \n",
|
| 493 |
+
"0 72.93 37.30 60.82 79.70 58.36 \n",
|
| 494 |
+
"1 72.69 36.32 60.34 79.57 57.51 \n",
|
| 495 |
+
"2 72.45 25.93 59.75 81.92 59.22 \n",
|
| 496 |
+
"3 64.56 17.59 57.60 77.65 55.89 \n",
|
| 497 |
+
"4 63.30 21.15 58.50 74.94 52.73 \n",
|
| 498 |
+
".. ... ... ... ... ... \n",
|
| 499 |
+
"258 68.82 38.13 59.43 83.45 65.70 \n",
|
| 500 |
+
"259 73.56 59.89 70.60 86.68 72.01 \n",
|
| 501 |
+
"260 72.53 66.19 63.72 80.10 61.01 \n",
|
| 502 |
+
"261 58.41 40.18 45.18 71.57 45.82 \n",
|
| 503 |
+
"262 72.14 54.06 63.01 82.34 62.97 \n",
|
| 504 |
+
"\n",
|
| 505 |
+
"[263 rows x 8 columns]"
|
| 506 |
+
]
|
| 507 |
+
},
|
| 508 |
+
"execution_count": 14,
|
| 509 |
+
"metadata": {},
|
| 510 |
+
"output_type": "execute_result"
|
| 511 |
+
}
|
| 512 |
+
],
|
| 513 |
+
"source": [
|
| 514 |
+
"new_df = df.drop([\"Date\", \"Average\"], axis=1).groupby(\"Model\").max().reset_index()\n",
|
| 515 |
+
"new_df"
|
| 516 |
+
]
|
| 517 |
+
},
|
| 518 |
+
{
|
| 519 |
+
"cell_type": "code",
|
| 520 |
+
"execution_count": 16,
|
| 521 |
+
"metadata": {},
|
| 522 |
+
"outputs": [
|
| 523 |
+
{
|
| 524 |
+
"data": {
|
| 525 |
+
"text/html": [
|
| 526 |
+
"<div>\n",
|
| 527 |
+
"<style scoped>\n",
|
| 528 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
| 529 |
+
" vertical-align: middle;\n",
|
| 530 |
+
" }\n",
|
| 531 |
+
"\n",
|
| 532 |
+
" .dataframe tbody tr th {\n",
|
| 533 |
+
" vertical-align: top;\n",
|
| 534 |
+
" }\n",
|
| 535 |
+
"\n",
|
| 536 |
+
" .dataframe thead th {\n",
|
| 537 |
+
" text-align: right;\n",
|
| 538 |
+
" }\n",
|
| 539 |
+
"</style>\n",
|
| 540 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
| 541 |
+
" <thead>\n",
|
| 542 |
+
" <tr style=\"text-align: right;\">\n",
|
| 543 |
+
" <th></th>\n",
|
| 544 |
+
" <th>Model</th>\n",
|
| 545 |
+
" <th>Date</th>\n",
|
| 546 |
+
" <th>Ifeval</th>\n",
|
| 547 |
+
" <th>Truthfulqa</th>\n",
|
| 548 |
+
" <th>Winogrande</th>\n",
|
| 549 |
+
" <th>Gsm8k</th>\n",
|
| 550 |
+
" <th>Mmlu</th>\n",
|
| 551 |
+
" <th>Hellaswag</th>\n",
|
| 552 |
+
" <th>Arc</th>\n",
|
| 553 |
+
" </tr>\n",
|
| 554 |
+
" </thead>\n",
|
| 555 |
+
" <tbody>\n",
|
| 556 |
+
" <tr>\n",
|
| 557 |
+
" <th>0</th>\n",
|
| 558 |
+
" <td>NousResearch_Nous-Hermes-2-Yi-34B_main</td>\n",
|
| 559 |
+
" <td>2024-03-04</td>\n",
|
| 560 |
+
" <td>39.00</td>\n",
|
| 561 |
+
" <td>61.44</td>\n",
|
| 562 |
+
" <td>80.58</td>\n",
|
| 563 |
+
" <td>67.93</td>\n",
|
| 564 |
+
" <td>76.24</td>\n",
|
| 565 |
+
" <td>83.79</td>\n",
|
| 566 |
+
" <td>68.00</td>\n",
|
| 567 |
+
" </tr>\n",
|
| 568 |
+
" <tr>\n",
|
| 569 |
+
" <th>1</th>\n",
|
| 570 |
+
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
|
| 571 |
+
" <td>2024-03-05</td>\n",
|
| 572 |
+
" <td>55.27</td>\n",
|
| 573 |
+
" <td>57.78</td>\n",
|
| 574 |
+
" <td>79.16</td>\n",
|
| 575 |
+
" <td>76.12</td>\n",
|
| 576 |
+
" <td>71.18</td>\n",
|
| 577 |
+
" <td>83.94</td>\n",
|
| 578 |
+
" <td>64.16</td>\n",
|
| 579 |
+
" </tr>\n",
|
| 580 |
+
" <tr>\n",
|
| 581 |
+
" <th>2</th>\n",
|
| 582 |
+
" <td>NousResearch_Nous-Hermes-2-Mixtral-8x7B-DPO_main</td>\n",
|
| 583 |
+
" <td>2024-03-02</td>\n",
|
| 584 |
+
" <td>59.33</td>\n",
|
| 585 |
+
" <td>64.76</td>\n",
|
| 586 |
+
" <td>78.53</td>\n",
|
| 587 |
+
" <td>62.17</td>\n",
|
| 588 |
+
" <td>71.96</td>\n",
|
| 589 |
+
" <td>85.42</td>\n",
|
| 590 |
+
" <td>70.82</td>\n",
|
| 591 |
+
" </tr>\n",
|
| 592 |
+
" <tr>\n",
|
| 593 |
+
" <th>3</th>\n",
|
| 594 |
+
" <td>mistralai_Mixtral-8x7B-Instruct-v0.1_main</td>\n",
|
| 595 |
+
" <td>2024-03-02</td>\n",
|
| 596 |
+
" <td>55.08</td>\n",
|
| 597 |
+
" <td>70.79</td>\n",
|
| 598 |
+
" <td>73.56</td>\n",
|
| 599 |
+
" <td>59.89</td>\n",
|
| 600 |
+
" <td>70.60</td>\n",
|
| 601 |
+
" <td>86.68</td>\n",
|
| 602 |
+
" <td>72.01</td>\n",
|
| 603 |
+
" </tr>\n",
|
| 604 |
+
" <tr>\n",
|
| 605 |
+
" <th>4</th>\n",
|
| 606 |
+
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
|
| 607 |
+
" <td>2024-03-04</td>\n",
|
| 608 |
+
" <td>55.27</td>\n",
|
| 609 |
+
" <td>57.78</td>\n",
|
| 610 |
+
" <td>79.16</td>\n",
|
| 611 |
+
" <td>76.12</td>\n",
|
| 612 |
+
" <td>71.18</td>\n",
|
| 613 |
+
" <td>83.94</td>\n",
|
| 614 |
+
" <td>64.16</td>\n",
|
| 615 |
+
" </tr>\n",
|
| 616 |
+
" <tr>\n",
|
| 617 |
+
" <th>...</th>\n",
|
| 618 |
+
" <td>...</td>\n",
|
| 619 |
+
" <td>...</td>\n",
|
| 620 |
+
" <td>...</td>\n",
|
| 621 |
+
" <td>...</td>\n",
|
| 622 |
+
" <td>...</td>\n",
|
| 623 |
+
" <td>...</td>\n",
|
| 624 |
+
" <td>...</td>\n",
|
| 625 |
+
" <td>...</td>\n",
|
| 626 |
+
" <td>...</td>\n",
|
| 627 |
+
" </tr>\n",
|
| 628 |
+
" <tr>\n",
|
| 629 |
+
" <th>269</th>\n",
|
| 630 |
+
" <td>HuggingFaceH4_starcoder2-15b-ift_v18.0</td>\n",
|
| 631 |
+
" <td>2024-03-10</td>\n",
|
| 632 |
+
" <td>21.63</td>\n",
|
| 633 |
+
" <td>NaN</td>\n",
|
| 634 |
+
" <td>NaN</td>\n",
|
| 635 |
+
" <td>0.83</td>\n",
|
| 636 |
+
" <td>NaN</td>\n",
|
| 637 |
+
" <td>NaN</td>\n",
|
| 638 |
+
" <td>NaN</td>\n",
|
| 639 |
+
" </tr>\n",
|
| 640 |
+
" <tr>\n",
|
| 641 |
+
" <th>270</th>\n",
|
| 642 |
+
" <td>HuggingFaceH4_mistral-7b-ift_v49.0</td>\n",
|
| 643 |
+
" <td>2024-03-07</td>\n",
|
| 644 |
+
" <td>20.15</td>\n",
|
| 645 |
+
" <td>NaN</td>\n",
|
| 646 |
+
" <td>NaN</td>\n",
|
| 647 |
+
" <td>0.00</td>\n",
|
| 648 |
+
" <td>NaN</td>\n",
|
| 649 |
+
" <td>NaN</td>\n",
|
| 650 |
+
" <td>NaN</td>\n",
|
| 651 |
+
" </tr>\n",
|
| 652 |
+
" <tr>\n",
|
| 653 |
+
" <th>271</th>\n",
|
| 654 |
+
" <td>HuggingFaceH4_starchat-beta_main</td>\n",
|
| 655 |
+
" <td>2024-03-12</td>\n",
|
| 656 |
+
" <td>8.13</td>\n",
|
| 657 |
+
" <td>NaN</td>\n",
|
| 658 |
+
" <td>NaN</td>\n",
|
| 659 |
+
" <td>NaN</td>\n",
|
| 660 |
+
" <td>NaN</td>\n",
|
| 661 |
+
" <td>NaN</td>\n",
|
| 662 |
+
" <td>NaN</td>\n",
|
| 663 |
+
" </tr>\n",
|
| 664 |
+
" <tr>\n",
|
| 665 |
+
" <th>272</th>\n",
|
| 666 |
+
" <td>HuggingFaceH4_starcoder2-15b-ift_v7.0</td>\n",
|
| 667 |
+
" <td>2024-03-10</td>\n",
|
| 668 |
+
" <td>12.57</td>\n",
|
| 669 |
+
" <td>NaN</td>\n",
|
| 670 |
+
" <td>NaN</td>\n",
|
| 671 |
+
" <td>3.18</td>\n",
|
| 672 |
" <td>NaN</td>\n",
|
| 673 |
" <td>NaN</td>\n",
|
| 674 |
" <td>NaN</td>\n",
|
| 675 |
+
" </tr>\n",
|
| 676 |
+
" <tr>\n",
|
| 677 |
+
" <th>273</th>\n",
|
| 678 |
+
" <td>HuggingFaceH4_zephyr-7b-beta-ift_v1.1</td>\n",
|
| 679 |
+
" <td>2024-03-13</td>\n",
|
| 680 |
+
" <td>9.43</td>\n",
|
| 681 |
" <td>NaN</td>\n",
|
| 682 |
" <td>NaN</td>\n",
|
| 683 |
+
" <td>0.00</td>\n",
|
| 684 |
" <td>NaN</td>\n",
|
| 685 |
" <td>NaN</td>\n",
|
| 686 |
" <td>NaN</td>\n",
|
| 687 |
" </tr>\n",
|
| 688 |
" </tbody>\n",
|
| 689 |
"</table>\n",
|
| 690 |
+
"<p>274 rows × 9 columns</p>\n",
|
| 691 |
"</div>"
|
| 692 |
],
|
| 693 |
"text/plain": [
|
| 694 |
+
" Model Date Ifeval \\\n",
|
| 695 |
+
"0 NousResearch_Nous-Hermes-2-Yi-34B_main 2024-03-04 39.00 \n",
|
| 696 |
+
"1 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-05 55.27 \n",
|
| 697 |
+
"2 NousResearch_Nous-Hermes-2-Mixtral-8x7B-DPO_main 2024-03-02 59.33 \n",
|
| 698 |
+
"3 mistralai_Mixtral-8x7B-Instruct-v0.1_main 2024-03-02 55.08 \n",
|
| 699 |
+
"4 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-04 55.27 \n",
|
| 700 |
+
".. ... ... ... \n",
|
| 701 |
+
"269 HuggingFaceH4_starcoder2-15b-ift_v18.0 2024-03-10 21.63 \n",
|
| 702 |
+
"270 HuggingFaceH4_mistral-7b-ift_v49.0 2024-03-07 20.15 \n",
|
| 703 |
+
"271 HuggingFaceH4_starchat-beta_main 2024-03-12 8.13 \n",
|
| 704 |
+
"272 HuggingFaceH4_starcoder2-15b-ift_v7.0 2024-03-10 12.57 \n",
|
| 705 |
+
"273 HuggingFaceH4_zephyr-7b-beta-ift_v1.1 2024-03-13 9.43 \n",
|
| 706 |
+
"\n",
|
| 707 |
+
" Truthfulqa Winogrande Gsm8k Mmlu Hellaswag Arc \n",
|
| 708 |
+
"0 61.44 80.58 67.93 76.24 83.79 68.00 \n",
|
| 709 |
+
"1 57.78 79.16 76.12 71.18 83.94 64.16 \n",
|
| 710 |
+
"2 64.76 78.53 62.17 71.96 85.42 70.82 \n",
|
| 711 |
+
"3 70.79 73.56 59.89 70.60 86.68 72.01 \n",
|
| 712 |
+
"4 57.78 79.16 76.12 71.18 83.94 64.16 \n",
|
| 713 |
+
".. ... ... ... ... ... ... \n",
|
| 714 |
+
"269 NaN NaN 0.83 NaN NaN NaN \n",
|
| 715 |
+
"270 NaN NaN 0.00 NaN NaN NaN \n",
|
| 716 |
+
"271 NaN NaN NaN NaN NaN NaN \n",
|
| 717 |
+
"272 NaN NaN 3.18 NaN NaN NaN \n",
|
| 718 |
+
"273 NaN NaN 0.00 NaN NaN NaN \n",
|
| 719 |
"\n",
|
| 720 |
+
"[274 rows x 9 columns]"
|
|
|
|
|
|
|
| 721 |
]
|
| 722 |
},
|
| 723 |
+
"execution_count": 16,
|
| 724 |
"metadata": {},
|
| 725 |
"output_type": "execute_result"
|
| 726 |
}
|
| 727 |
],
|
| 728 |
"source": [
|
| 729 |
+
"df[[\"Model\", \"Date\"]].merge(new_df, on=\"Model\", how=\"left\")"
|
| 730 |
]
|
| 731 |
},
|
| 732 |
{
|