File size: 20,479 Bytes
599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8ea545e 599688f 8ea545e 8a31745 8ea545e 8a31745 599688f 8ea545e 599688f 8ea545e 8a31745 599688f 8a31745 8ea545e 599688f 8ea545e 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 599688f 8a31745 8ea545e 8a31745 8ea545e 8a31745 8ea545e 8a31745 8ea545e 8a31745 8ea545e 599688f 8ea545e 599688f 8ea545e 599688f 8ea545e 599688f 8ea545e 599688f 8ea545e 599688f 8ea545e 599688f 8ea545e 599688f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 |
{
"cells": [
{
"cell_type": "code",
"execution_count": 34,
"metadata": {},
"outputs": [],
"source": [
"import json\n",
"from pathlib import Path\n",
"\n",
"import gradio as gr\n",
"import pandas as pd\n",
"\n",
"TITLE = \"\"\"<h1 align=\"center\" id=\"space-title\">LLM Leaderboard for H4 Models</h1>\"\"\"\n",
"\n",
"DESCRIPTION = f\"\"\"\n",
"Evaluation of H4 and community models across a diverse range of benchmarks from [LightEval](https://github.com/huggingface/lighteval). All scores are reported as accuracy.\n",
"\"\"\"\n",
"\n",
"BENCHMARKS_TO_SKIP = [\"math\", \"mini_math\"]\n",
"\n",
"\n",
"def get_leaderboard_df(agg : str = \"max\"):\n",
" filepaths = list(Path(\"eval_results\").rglob(\"*.json\"))\n",
"\n",
" # Parse filepaths to get unique models\n",
" models = set()\n",
" for filepath in filepaths:\n",
" path_parts = Path(filepath).parts\n",
" model_revision = \"_\".join(path_parts[1:4])\n",
" models.add(model_revision)\n",
"\n",
" # Initialize DataFrame\n",
" df = pd.DataFrame(index=list(models))\n",
"\n",
" # Extract data from each file and populate the DataFrame\n",
" for filepath in filepaths:\n",
" path_parts = Path(filepath).parts\n",
" date = filepath.stem.split(\"_\")[-1][:-3]\n",
" model_revision = \"_\".join(path_parts[1:4]) + \"_\" + date\n",
" task = path_parts[4]\n",
" df.loc[model_revision, \"Date\"] = date\n",
"\n",
" with open(filepath, \"r\") as file:\n",
" data = json.load(file)\n",
" first_result_key = next(iter(data[\"results\"])) # gets the first key in 'results'\n",
" # Skip benchmarks that we don't want to include in the leaderboard\n",
" if task.lower() in BENCHMARKS_TO_SKIP:\n",
" continue\n",
" # TruthfulQA has two metrics, so we need to pick the `mc2` one that's reported on the leaderboard\n",
" if task.lower() == \"truthfulqa\":\n",
" value = data[\"results\"][first_result_key][\"truthfulqa_mc2\"]\n",
" # IFEval has several metrics but we report just the prompt-loose-acc one\n",
" elif task.lower() == \"ifeval\":\n",
" value = data[\"results\"][first_result_key][\"prompt_level_loose_acc\"]\n",
" # MMLU has several metrics but we report just the average one\n",
" elif task.lower() == \"mmlu\":\n",
" value = [v[\"acc\"] for k, v in data[\"results\"].items() if \"_average\" in k.lower()][0]\n",
" # HellaSwag and ARC reports acc_norm\n",
" elif task.lower() in [\"hellaswag\", \"arc\"]:\n",
" value = data[\"results\"][first_result_key][\"acc_norm\"]\n",
" # BBH has several metrics but we report just the average one\n",
" elif task.lower() == \"bbh\":\n",
" if \"all\" in data[\"results\"]:\n",
" value = data[\"results\"][\"all\"][\"acc\"]\n",
" else:\n",
" value = -100\n",
" # AGIEval reports acc_norm\n",
" elif task.lower() == \"agieval\":\n",
" value = data[\"results\"][\"all\"][\"acc_norm\"]\n",
" # MATH reports qem\n",
" elif task.lower() in [\"math\", \"math_v2\", \"aimo_kaggle\"]:\n",
" value = data[\"results\"][\"all\"][\"qem\"]\n",
" else:\n",
" first_metric_key = next(\n",
" iter(data[\"results\"][first_result_key])\n",
" ) # gets the first key in the first result\n",
" value = data[\"results\"][first_result_key][first_metric_key] # gets the value of the first metric\n",
"\n",
" # For mini_math we report 5 metrics, one for each level and store each one as a separate row in the dataframe\n",
" if task.lower() in [\"mini_math_v2\"]:\n",
" for k, v in data[\"results\"].items():\n",
" if k != \"all\":\n",
" level = k.split(\"|\")[1].split(\":\")[-1]\n",
" value = v[\"qem\"]\n",
" df.loc[model_revision, f\"{task}_{level}\"] = value\n",
" # For kaggle_pot we report N metrics, one for each prompt and store each one as a separate row in the dataframe\n",
" elif task.lower() in [\"aimo_kaggle_medium_pot\"]:\n",
" for k, v in data[\"results\"].items():\n",
" if k != \"all\" and \"_average\" not in k:\n",
" version = k.split(\"|\")[1].split(\":\")[-1]\n",
" value = v[\"qem\"] if \"qem\" in v else v[\"score\"]\n",
" df.loc[model_revision, f\"{task}_{version}\"] = value\n",
" # For kaggle_pot we report N metrics, one for each prompt and store each one as a separate row in the dataframe\n",
" elif task.lower() in [\"aimo_kaggle_hard_pot\"]:\n",
" for k, v in data[\"results\"].items():\n",
" if k != \"all\" and \"_average\" not in k:\n",
" version = k.split(\"|\")[1].split(\":\")[-1]\n",
" value = v[\"qem\"] if \"qem\" in v else v[\"score\"]\n",
" df.loc[model_revision, f\"{task}_{version}\"] = value\n",
" # For kaggle_tora we report accuracy, so need to divide by 100\n",
" elif task.lower() in [\n",
" \"aimo_tora_eval_kaggle_medium\",\n",
" \"aimo_tora_eval_kaggle_hard\",\n",
" \"aimo_kaggle_fast_eval_hard\",\n",
" \"aimo_kaggle_tora_medium\",\n",
" \"aimo_kaggle_tora_hard\",\n",
" \"aimo_kaggle_tora_medium_extended\",\n",
" \"aimo_kaggle_tora_hard_extended\",\n",
" ]:\n",
" for k, v in data[\"results\"].items():\n",
" value = float(v[\"qem\"]) / 100.0\n",
" df.loc[model_revision, f\"{task}\"] = value\n",
" # For AlpacaEval we report base winrate and lenght corrected one\n",
" elif task.lower() == \"alpaca_eval\":\n",
" value = data[\"results\"][first_result_key][\"win_rate\"]\n",
" df.loc[model_revision, \"Alpaca_eval\"] = value / 100.0\n",
" value = data[\"results\"][first_result_key][\"length_controlled_winrate\"]\n",
" df.loc[model_revision, \"Alpaca_eval_lc\"] = value / 100.0\n",
" else:\n",
" df.loc[model_revision, task] = float(value)\n",
"\n",
" # Drop rows where every entry is NaN\n",
" df = df.dropna(how=\"all\", axis=0, subset=[c for c in df.columns if c != \"Date\"])\n",
"\n",
" # Trim minimath column names\n",
" df.columns = [c.replace(\"_level_\", \"_l\") for c in df.columns]\n",
"\n",
" # Trim AIMO column names\n",
" df.columns = [c.replace(\"aimo_\", \"\") for c in df.columns]\n",
"\n",
" df.insert(loc=0, column=\"Average\", value=df.mean(axis=1, numeric_only=True))\n",
"\n",
" # Convert all values to percentage\n",
" df[df.select_dtypes(include=[\"number\"]).columns] *= 100.0\n",
" df = df.sort_values(by=[\"Average\"], ascending=False)\n",
" df = df.reset_index().rename(columns={\"index\": \"Model\"}).round(2)\n",
" # Strip off date from model name\n",
" df[\"Model\"] = df[\"Model\"].apply(lambda x: x.rsplit(\"_\", 1)[0])\n",
"\n",
" # Drop date and aggregate results by model name\n",
" df = df.drop(\"Date\", axis=1).groupby(\"Model\").agg(agg).reset_index()\n",
"\n",
" return df"
]
},
{
"cell_type": "code",
"execution_count": 41,
"metadata": {},
"outputs": [],
"source": [
"df = get_leaderboard_df(agg='mean')"
]
},
{
"cell_type": "code",
"execution_count": 37,
"metadata": {},
"outputs": [],
"source": [
"# df"
]
},
{
"cell_type": "code",
"execution_count": 40,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Model</th>\n",
" <th>Average</th>\n",
" <th>kaggle_tora_medium_extended</th>\n",
" <th>kaggle_tora_hard_extended</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>1741</th>\n",
" <td>AI-MO_deepseek-math-7b-sft_aimo_v38.15.gptq-8bits</td>\n",
" <td>28.89</td>\n",
" <td>61.45</td>\n",
" <td>28.89</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Model Average \\\n",
"1741 AI-MO_deepseek-math-7b-sft_aimo_v38.15.gptq-8bits 28.89 \n",
"\n",
" kaggle_tora_medium_extended kaggle_tora_hard_extended \n",
"1741 61.45 28.89 "
]
},
"execution_count": 40,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.query(\"Model == 'AI-MO_deepseek-math-7b-sft_aimo_v38.15.gptq-8bits'\").dropna(axis=1, how=\"all\")"
]
},
{
"cell_type": "code",
"execution_count": 38,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Model</th>\n",
" <th>Average</th>\n",
" <th>kaggle_tora_medium_extended</th>\n",
" <th>kaggle_tora_hard_extended</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>1741</th>\n",
" <td>AI-MO_deepseek-math-7b-sft_aimo_v38.15.gptq-8bits</td>\n",
" <td>65.06</td>\n",
" <td>65.06</td>\n",
" <td>32.22</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Model Average \\\n",
"1741 AI-MO_deepseek-math-7b-sft_aimo_v38.15.gptq-8bits 65.06 \n",
"\n",
" kaggle_tora_medium_extended kaggle_tora_hard_extended \n",
"1741 65.06 32.22 "
]
},
"execution_count": 38,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.query(\"Model == 'AI-MO_deepseek-math-7b-sft_aimo_v38.15.gptq-8bits'\").dropna(axis=1, how=\"all\")"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Model</th>\n",
" <th>Date</th>\n",
" <th>Ifeval</th>\n",
" <th>Truthfulqa</th>\n",
" <th>Winogrande</th>\n",
" <th>Gsm8k</th>\n",
" <th>Mmlu</th>\n",
" <th>Hellaswag</th>\n",
" <th>Arc</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>NousResearch_Nous-Hermes-2-Yi-34B_main</td>\n",
" <td>2024-03-04</td>\n",
" <td>39.00</td>\n",
" <td>61.44</td>\n",
" <td>80.58</td>\n",
" <td>67.93</td>\n",
" <td>76.24</td>\n",
" <td>83.79</td>\n",
" <td>68.00</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
" <td>2024-03-05</td>\n",
" <td>55.27</td>\n",
" <td>57.78</td>\n",
" <td>79.16</td>\n",
" <td>76.12</td>\n",
" <td>71.18</td>\n",
" <td>83.94</td>\n",
" <td>64.16</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>NousResearch_Nous-Hermes-2-Mixtral-8x7B-DPO_main</td>\n",
" <td>2024-03-02</td>\n",
" <td>59.33</td>\n",
" <td>64.76</td>\n",
" <td>78.53</td>\n",
" <td>62.17</td>\n",
" <td>71.96</td>\n",
" <td>85.42</td>\n",
" <td>70.82</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>mistralai_Mixtral-8x7B-Instruct-v0.1_main</td>\n",
" <td>2024-03-02</td>\n",
" <td>55.08</td>\n",
" <td>70.79</td>\n",
" <td>73.56</td>\n",
" <td>59.89</td>\n",
" <td>70.60</td>\n",
" <td>86.68</td>\n",
" <td>72.01</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>deepseek-ai_deepseek-llm-67b-chat_main</td>\n",
" <td>2024-03-04</td>\n",
" <td>55.27</td>\n",
" <td>57.78</td>\n",
" <td>79.16</td>\n",
" <td>76.12</td>\n",
" <td>71.18</td>\n",
" <td>83.94</td>\n",
" <td>64.16</td>\n",
" </tr>\n",
" <tr>\n",
" <th>...</th>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" </tr>\n",
" <tr>\n",
" <th>269</th>\n",
" <td>HuggingFaceH4_starcoder2-15b-ift_v18.0</td>\n",
" <td>2024-03-10</td>\n",
" <td>21.63</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>0.83</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" </tr>\n",
" <tr>\n",
" <th>270</th>\n",
" <td>HuggingFaceH4_mistral-7b-ift_v49.0</td>\n",
" <td>2024-03-07</td>\n",
" <td>20.15</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>0.00</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" </tr>\n",
" <tr>\n",
" <th>271</th>\n",
" <td>HuggingFaceH4_starchat-beta_main</td>\n",
" <td>2024-03-12</td>\n",
" <td>8.13</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" </tr>\n",
" <tr>\n",
" <th>272</th>\n",
" <td>HuggingFaceH4_starcoder2-15b-ift_v7.0</td>\n",
" <td>2024-03-10</td>\n",
" <td>12.57</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>3.18</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" </tr>\n",
" <tr>\n",
" <th>273</th>\n",
" <td>HuggingFaceH4_zephyr-7b-beta-ift_v1.1</td>\n",
" <td>2024-03-13</td>\n",
" <td>9.43</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>0.00</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>274 rows × 9 columns</p>\n",
"</div>"
],
"text/plain": [
" Model Date Ifeval \\\n",
"0 NousResearch_Nous-Hermes-2-Yi-34B_main 2024-03-04 39.00 \n",
"1 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-05 55.27 \n",
"2 NousResearch_Nous-Hermes-2-Mixtral-8x7B-DPO_main 2024-03-02 59.33 \n",
"3 mistralai_Mixtral-8x7B-Instruct-v0.1_main 2024-03-02 55.08 \n",
"4 deepseek-ai_deepseek-llm-67b-chat_main 2024-03-04 55.27 \n",
".. ... ... ... \n",
"269 HuggingFaceH4_starcoder2-15b-ift_v18.0 2024-03-10 21.63 \n",
"270 HuggingFaceH4_mistral-7b-ift_v49.0 2024-03-07 20.15 \n",
"271 HuggingFaceH4_starchat-beta_main 2024-03-12 8.13 \n",
"272 HuggingFaceH4_starcoder2-15b-ift_v7.0 2024-03-10 12.57 \n",
"273 HuggingFaceH4_zephyr-7b-beta-ift_v1.1 2024-03-13 9.43 \n",
"\n",
" Truthfulqa Winogrande Gsm8k Mmlu Hellaswag Arc \n",
"0 61.44 80.58 67.93 76.24 83.79 68.00 \n",
"1 57.78 79.16 76.12 71.18 83.94 64.16 \n",
"2 64.76 78.53 62.17 71.96 85.42 70.82 \n",
"3 70.79 73.56 59.89 70.60 86.68 72.01 \n",
"4 57.78 79.16 76.12 71.18 83.94 64.16 \n",
".. ... ... ... ... ... ... \n",
"269 NaN NaN 0.83 NaN NaN NaN \n",
"270 NaN NaN 0.00 NaN NaN NaN \n",
"271 NaN NaN NaN NaN NaN NaN \n",
"272 NaN NaN 3.18 NaN NaN NaN \n",
"273 NaN NaN 0.00 NaN NaN NaN \n",
"\n",
"[274 rows x 9 columns]"
]
},
"execution_count": 16,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df[[\"Model\", \"Date\"]].merge(new_df, on=\"Model\", how=\"left\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "hf",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.6"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
|