{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fetch the data from the hub"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import itertools\n",
    "import pandas as pd\n",
    "from concurrent.futures import ThreadPoolExecutor\n",
    "from tqdm import tqdm\n",
    "import itertools\n",
    "import huggingface_hub\n",
    "from tensorboard.backend.event_processing.event_accumulator import EventAccumulator\n",
    "from huggingface_hub.utils import EntryNotFoundError"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "def step_element_match(step_to_check, step_element):\n",
    "    step_element = step_element.strip().replace(\" \", \"\")\n",
    "    if \"-\" in step_element:\n",
    "        a, b = step_element.split(\"-\")\n",
    "        c = None\n",
    "        if \"%\" in b:\n",
    "            b, c = b.split(\"%\")\n",
    "        return (int(a) <= step_to_check <= int(b) and\n",
    "                (c is None or (step_to_check - int(a)) % int(c) == 0))\n",
    "    elif \"%\" in step_element:\n",
    "        return step_to_check % int(step_element[1:]) == 0\n",
    "    else:\n",
    "        return step_to_check == int(step_element)\n",
    "    \n",
    "def fetch_run_results_simple(repo_name, runs_to_fetch, steps_to_fetch, prefix, agg_score_columns, column_name,\n",
    "                             seed_merge_method, oauth_token=None, prefix_file=None):\n",
    "    if not runs_to_fetch:\n",
    "        return\n",
    "\n",
    "    def fetch_run_files(run_to_fetch):\n",
    "        def filename_to_steps_timestamp(fn):\n",
    "            step, ts = fn.split(\"_events.out.tfevents.\")\n",
    "            return int(step[-7:]), int(ts[:ts.index(\".\")])\n",
    "\n",
    "        run_to_fetch += \"_e\"\n",
    "        try:\n",
    "            eval_repo_file_names = [f.path for f in\n",
    "                                    huggingface_hub.list_repo_tree(repo_name, run_to_fetch, expand=False,\n",
    "                                                                   token=oauth_token) if\n",
    "                                    \"_events.out.tfevents\" in f.path]\n",
    "        except EntryNotFoundError:\n",
    "            return []\n",
    "\n",
    "        eval_files = [os.path.relpath(f, run_to_fetch) for f in eval_repo_file_names]\n",
    "        timestamps = {}\n",
    "        for fn in eval_files:\n",
    "            steps, ts = filename_to_steps_timestamp(fn)\n",
    "            if steps not in timestamps or timestamps[steps][0] < ts:\n",
    "                timestamps[steps] = ts, fn\n",
    "\n",
    "        results = []\n",
    "        for eval_file, repofile in zip(eval_files, eval_repo_file_names):\n",
    "            steps, ts = filename_to_steps_timestamp(eval_file)\n",
    "            if not any(step_element_match(steps, step_el) for step_el in steps_to_fetch.split(\",\")):\n",
    "                continue\n",
    "            if timestamps[steps][1] == eval_file:\n",
    "                results.append((run_to_fetch, steps, repofile))\n",
    "        return results\n",
    "\n",
    "    def load_run_file(data):\n",
    "        run_to_fetch, steps, repofile = data\n",
    "        loader = EventAccumulator(huggingface_hub.hf_hub_download(repo_name, repofile, token=oauth_token))\n",
    "        loader.Reload()\n",
    "        runname = run_to_fetch.removeprefix(prefix).removesuffix(\"-_e\")\n",
    "        column_names = [\"runname\", \"seed\", \"steps\", \"agg_score\"]\n",
    "        column_values = [runname, 0, steps, 0.0]\n",
    "\n",
    "        for tag in loader.Tags()['scalars']:\n",
    "            if not \"stderr\" in tag and tag.split('/')[0] == 'e':\n",
    "                event_list = loader.Scalars(tag)\n",
    "                tag = tag.split('/')\n",
    "                column_names.append(f\"{tag[1]}/{tag[2]}\")\n",
    "                column_values.append(event_list[-1].value)\n",
    "\n",
    "        return pd.DataFrame([column_values], columns=column_names)\n",
    "\n",
    "    with ThreadPoolExecutor() as pool:\n",
    "        run_files = list(itertools.chain.from_iterable(\n",
    "            tqdm(pool.map(fetch_run_files, runs_to_fetch), total=len(runs_to_fetch), desc=\"Fetching datafiles...\")))\n",
    "        df = pd.concat(tqdm(pool.map(load_run_file, run_files), total=len(run_files), desc=\"Loading evals data...\"))\n",
    "\n",
    "    cols_to_avg = [col for col in agg_score_columns if col in df.columns]\n",
    "    if cols_to_avg:\n",
    "        df['agg_score'] = df[cols_to_avg].mean(axis=1)\n",
    "\n",
    "    prefix_file = prefix_file + \"_\" if prefix_file else \"\"\n",
    "    df.to_csv(f\"{prefix_file}{repo_name.split('/')[-1]}_metrics.csv\", index=False)\n",
    "    print(f\"Metrics saved to {repo_name.split('/')[-1]}_metrics.csv\")\n",
    "\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Fetching datafiles...: 100%|██████████| 1/1 [00:02<00:00,  2.94s/it]\n",
      "Loading evals data...: 100%|██████████| 82/82 [00:15<00:00,  5.37it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Metrics saved to loubna-edu_fw_ablations_metrics.csv\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "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>runname</th>\n",
       "      <th>seed</th>\n",
       "      <th>steps</th>\n",
       "      <th>agg_score</th>\n",
       "      <th>commonsense_qa/acc</th>\n",
       "      <th>commonsense_qa/acc_norm</th>\n",
       "      <th>hellaswag/acc</th>\n",
       "      <th>hellaswag/acc_norm</th>\n",
       "      <th>openbookqa/acc</th>\n",
       "      <th>openbookqa/acc_norm</th>\n",
       "      <th>...</th>\n",
       "      <th>siqa/acc</th>\n",
       "      <th>siqa/acc_norm</th>\n",
       "      <th>winogrande/acc</th>\n",
       "      <th>winogrande/acc_norm</th>\n",
       "      <th>all/acc</th>\n",
       "      <th>all/acc_norm</th>\n",
       "      <th>arc/acc</th>\n",
       "      <th>arc/acc_norm</th>\n",
       "      <th>mmlu/acc</th>\n",
       "      <th>mmlu/acc_norm</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>2000</td>\n",
       "      <td>0.390326</td>\n",
       "      <td>0.284</td>\n",
       "      <td>0.283</td>\n",
       "      <td>0.314</td>\n",
       "      <td>0.325</td>\n",
       "      <td>0.164</td>\n",
       "      <td>0.296</td>\n",
       "      <td>...</td>\n",
       "      <td>0.362</td>\n",
       "      <td>0.406</td>\n",
       "      <td>0.511</td>\n",
       "      <td>0.511</td>\n",
       "      <td>0.279674</td>\n",
       "      <td>0.299162</td>\n",
       "      <td>0.3795</td>\n",
       "      <td>0.3850</td>\n",
       "      <td>0.265997</td>\n",
       "      <td>0.284605</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>4000</td>\n",
       "      <td>0.414680</td>\n",
       "      <td>0.322</td>\n",
       "      <td>0.307</td>\n",
       "      <td>0.343</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.196</td>\n",
       "      <td>0.320</td>\n",
       "      <td>...</td>\n",
       "      <td>0.371</td>\n",
       "      <td>0.388</td>\n",
       "      <td>0.518</td>\n",
       "      <td>0.495</td>\n",
       "      <td>0.290613</td>\n",
       "      <td>0.312593</td>\n",
       "      <td>0.4215</td>\n",
       "      <td>0.4285</td>\n",
       "      <td>0.274401</td>\n",
       "      <td>0.295939</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>6000</td>\n",
       "      <td>0.428390</td>\n",
       "      <td>0.319</td>\n",
       "      <td>0.311</td>\n",
       "      <td>0.372</td>\n",
       "      <td>0.431</td>\n",
       "      <td>0.202</td>\n",
       "      <td>0.352</td>\n",
       "      <td>...</td>\n",
       "      <td>0.373</td>\n",
       "      <td>0.392</td>\n",
       "      <td>0.520</td>\n",
       "      <td>0.519</td>\n",
       "      <td>0.303980</td>\n",
       "      <td>0.323323</td>\n",
       "      <td>0.4315</td>\n",
       "      <td>0.4460</td>\n",
       "      <td>0.288591</td>\n",
       "      <td>0.306123</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>0.443615</td>\n",
       "      <td>0.340</td>\n",
       "      <td>0.311</td>\n",
       "      <td>0.379</td>\n",
       "      <td>0.463</td>\n",
       "      <td>0.204</td>\n",
       "      <td>0.360</td>\n",
       "      <td>...</td>\n",
       "      <td>0.384</td>\n",
       "      <td>0.404</td>\n",
       "      <td>0.517</td>\n",
       "      <td>0.517</td>\n",
       "      <td>0.315148</td>\n",
       "      <td>0.333284</td>\n",
       "      <td>0.4630</td>\n",
       "      <td>0.4790</td>\n",
       "      <td>0.299186</td>\n",
       "      <td>0.314921</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>10000</td>\n",
       "      <td>0.441457</td>\n",
       "      <td>0.346</td>\n",
       "      <td>0.317</td>\n",
       "      <td>0.390</td>\n",
       "      <td>0.454</td>\n",
       "      <td>0.222</td>\n",
       "      <td>0.364</td>\n",
       "      <td>...</td>\n",
       "      <td>0.366</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.514</td>\n",
       "      <td>0.506</td>\n",
       "      <td>0.318935</td>\n",
       "      <td>0.335419</td>\n",
       "      <td>0.4890</td>\n",
       "      <td>0.4820</td>\n",
       "      <td>0.302189</td>\n",
       "      <td>0.317653</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",
       "      <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",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>160000</td>\n",
       "      <td>0.507129</td>\n",
       "      <td>0.430</td>\n",
       "      <td>0.359</td>\n",
       "      <td>0.473</td>\n",
       "      <td>0.593</td>\n",
       "      <td>0.282</td>\n",
       "      <td>0.418</td>\n",
       "      <td>...</td>\n",
       "      <td>0.392</td>\n",
       "      <td>0.402</td>\n",
       "      <td>0.576</td>\n",
       "      <td>0.575</td>\n",
       "      <td>0.369137</td>\n",
       "      <td>0.393898</td>\n",
       "      <td>0.5670</td>\n",
       "      <td>0.5725</td>\n",
       "      <td>0.350226</td>\n",
       "      <td>0.374533</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>162000</td>\n",
       "      <td>0.509118</td>\n",
       "      <td>0.416</td>\n",
       "      <td>0.367</td>\n",
       "      <td>0.474</td>\n",
       "      <td>0.592</td>\n",
       "      <td>0.288</td>\n",
       "      <td>0.408</td>\n",
       "      <td>...</td>\n",
       "      <td>0.390</td>\n",
       "      <td>0.409</td>\n",
       "      <td>0.572</td>\n",
       "      <td>0.577</td>\n",
       "      <td>0.367420</td>\n",
       "      <td>0.392861</td>\n",
       "      <td>0.5720</td>\n",
       "      <td>0.5780</td>\n",
       "      <td>0.348268</td>\n",
       "      <td>0.372947</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>164000</td>\n",
       "      <td>0.507843</td>\n",
       "      <td>0.416</td>\n",
       "      <td>0.365</td>\n",
       "      <td>0.467</td>\n",
       "      <td>0.591</td>\n",
       "      <td>0.276</td>\n",
       "      <td>0.408</td>\n",
       "      <td>...</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.406</td>\n",
       "      <td>0.576</td>\n",
       "      <td>0.580</td>\n",
       "      <td>0.368319</td>\n",
       "      <td>0.392000</td>\n",
       "      <td>0.5635</td>\n",
       "      <td>0.5715</td>\n",
       "      <td>0.349943</td>\n",
       "      <td>0.372246</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>166000</td>\n",
       "      <td>0.508308</td>\n",
       "      <td>0.415</td>\n",
       "      <td>0.364</td>\n",
       "      <td>0.472</td>\n",
       "      <td>0.593</td>\n",
       "      <td>0.282</td>\n",
       "      <td>0.414</td>\n",
       "      <td>...</td>\n",
       "      <td>0.401</td>\n",
       "      <td>0.408</td>\n",
       "      <td>0.575</td>\n",
       "      <td>0.570</td>\n",
       "      <td>0.370593</td>\n",
       "      <td>0.393176</td>\n",
       "      <td>0.5640</td>\n",
       "      <td>0.5760</td>\n",
       "      <td>0.352203</td>\n",
       "      <td>0.373463</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>edu_fineweb_350b_tokens-seed-1</td>\n",
       "      <td>0</td>\n",
       "      <td>167000</td>\n",
       "      <td>0.509494</td>\n",
       "      <td>0.429</td>\n",
       "      <td>0.362</td>\n",
       "      <td>0.472</td>\n",
       "      <td>0.597</td>\n",
       "      <td>0.290</td>\n",
       "      <td>0.418</td>\n",
       "      <td>...</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.404</td>\n",
       "      <td>0.582</td>\n",
       "      <td>0.578</td>\n",
       "      <td>0.369666</td>\n",
       "      <td>0.394136</td>\n",
       "      <td>0.5670</td>\n",
       "      <td>0.5735</td>\n",
       "      <td>0.350671</td>\n",
       "      <td>0.374453</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>82 rows × 22 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                           runname  seed   steps  agg_score  \\\n",
       "0   edu_fineweb_350b_tokens-seed-1     0    2000   0.390326   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0    4000   0.414680   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0    6000   0.428390   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0    8000   0.443615   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0   10000   0.441457   \n",
       "..                             ...   ...     ...        ...   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0  160000   0.507129   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0  162000   0.509118   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0  164000   0.507843   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0  166000   0.508308   \n",
       "0   edu_fineweb_350b_tokens-seed-1     0  167000   0.509494   \n",
       "\n",
       "    commonsense_qa/acc  commonsense_qa/acc_norm  hellaswag/acc  \\\n",
       "0                0.284                    0.283          0.314   \n",
       "0                0.322                    0.307          0.343   \n",
       "0                0.319                    0.311          0.372   \n",
       "0                0.340                    0.311          0.379   \n",
       "0                0.346                    0.317          0.390   \n",
       "..                 ...                      ...            ...   \n",
       "0                0.430                    0.359          0.473   \n",
       "0                0.416                    0.367          0.474   \n",
       "0                0.416                    0.365          0.467   \n",
       "0                0.415                    0.364          0.472   \n",
       "0                0.429                    0.362          0.472   \n",
       "\n",
       "    hellaswag/acc_norm  openbookqa/acc  openbookqa/acc_norm  ...  siqa/acc  \\\n",
       "0                0.325           0.164                0.296  ...     0.362   \n",
       "0                0.395           0.196                0.320  ...     0.371   \n",
       "0                0.431           0.202                0.352  ...     0.373   \n",
       "0                0.463           0.204                0.360  ...     0.384   \n",
       "0                0.454           0.222                0.364  ...     0.366   \n",
       "..                 ...             ...                  ...  ...       ...   \n",
       "0                0.593           0.282                0.418  ...     0.392   \n",
       "0                0.592           0.288                0.408  ...     0.390   \n",
       "0                0.591           0.276                0.408  ...     0.395   \n",
       "0                0.593           0.282                0.414  ...     0.401   \n",
       "0                0.597           0.290                0.418  ...     0.395   \n",
       "\n",
       "    siqa/acc_norm  winogrande/acc  winogrande/acc_norm   all/acc  \\\n",
       "0           0.406           0.511                0.511  0.279674   \n",
       "0           0.388           0.518                0.495  0.290613   \n",
       "0           0.392           0.520                0.519  0.303980   \n",
       "0           0.404           0.517                0.517  0.315148   \n",
       "0           0.395           0.514                0.506  0.318935   \n",
       "..            ...             ...                  ...       ...   \n",
       "0           0.402           0.576                0.575  0.369137   \n",
       "0           0.409           0.572                0.577  0.367420   \n",
       "0           0.406           0.576                0.580  0.368319   \n",
       "0           0.408           0.575                0.570  0.370593   \n",
       "0           0.404           0.582                0.578  0.369666   \n",
       "\n",
       "    all/acc_norm  arc/acc  arc/acc_norm  mmlu/acc  mmlu/acc_norm  \n",
       "0       0.299162   0.3795        0.3850  0.265997       0.284605  \n",
       "0       0.312593   0.4215        0.4285  0.274401       0.295939  \n",
       "0       0.323323   0.4315        0.4460  0.288591       0.306123  \n",
       "0       0.333284   0.4630        0.4790  0.299186       0.314921  \n",
       "0       0.335419   0.4890        0.4820  0.302189       0.317653  \n",
       "..           ...      ...           ...       ...            ...  \n",
       "0       0.393898   0.5670        0.5725  0.350226       0.374533  \n",
       "0       0.392861   0.5720        0.5780  0.348268       0.372947  \n",
       "0       0.392000   0.5635        0.5715  0.349943       0.372246  \n",
       "0       0.393176   0.5640        0.5760  0.352203       0.373463  \n",
       "0       0.394136   0.5670        0.5735  0.350671       0.374453  \n",
       "\n",
       "[82 rows x 22 columns]"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "token = os.getenv(\"HF_TOKEN\")\n",
    "repo_name = \"HuggingFaceTB/loubna-edu_fw_ablations\"\n",
    "runs_to_fetch = [\"tb/edu_fw_ablations-1p82G-edu_fineweb_350b_tokens-seed-1-\"]\n",
    "steps_to_fetch = \"%1000\"\n",
    "prefix = \"tb/edu_fw_ablations-1p82G-\"\n",
    "metrics = ['commonsense_qa/acc_norm', 'hellaswag/acc_norm', 'openbookqa/acc_norm', 'piqa/acc_norm',\n",
    "                   'siqa/acc_norm', 'winogrande/acc_norm', 'arc/acc_norm', 'mmlu/acc_norm']\n",
    "agg_score_columns = metrics\n",
    "column_name = \"agg_score\"\n",
    "seed_merge_method = \"mean\"\n",
    "oauth_token = token\n",
    "\n",
    "# runs_to_fetch = [prefix + run for run in runs_to_fetch]\n",
    "fetch_run_results_simple(repo_name, runs_to_fetch, steps_to_fetch, prefix, agg_score_columns, column_name, seed_merge_method, oauth_token=token)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot the data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load csvs for FW and FW-Edu"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "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>runname</th>\n",
       "      <th>steps</th>\n",
       "      <th>agg_score</th>\n",
       "      <th>commonsense_qa/acc</th>\n",
       "      <th>commonsense_qa/acc_norm</th>\n",
       "      <th>hellaswag/acc</th>\n",
       "      <th>hellaswag/acc_norm</th>\n",
       "      <th>openbookqa/acc</th>\n",
       "      <th>openbookqa/acc_norm</th>\n",
       "      <th>piqa/acc</th>\n",
       "      <th>piqa/acc_norm</th>\n",
       "      <th>siqa/acc</th>\n",
       "      <th>siqa/acc_norm</th>\n",
       "      <th>winogrande/acc</th>\n",
       "      <th>winogrande/acc_norm</th>\n",
       "      <th>arc/acc</th>\n",
       "      <th>arc/acc_norm</th>\n",
       "      <th>mmlu/acc</th>\n",
       "      <th>mmlu/acc_norm</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>2000</td>\n",
       "      <td>0.390326</td>\n",
       "      <td>0.284</td>\n",
       "      <td>0.283</td>\n",
       "      <td>0.314</td>\n",
       "      <td>0.325</td>\n",
       "      <td>0.164</td>\n",
       "      <td>0.296</td>\n",
       "      <td>0.623</td>\n",
       "      <td>0.632</td>\n",
       "      <td>0.362</td>\n",
       "      <td>0.406</td>\n",
       "      <td>0.511</td>\n",
       "      <td>0.511</td>\n",
       "      <td>0.3795</td>\n",
       "      <td>0.3850</td>\n",
       "      <td>0.265997</td>\n",
       "      <td>0.284605</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>4000</td>\n",
       "      <td>0.414680</td>\n",
       "      <td>0.322</td>\n",
       "      <td>0.307</td>\n",
       "      <td>0.343</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.196</td>\n",
       "      <td>0.320</td>\n",
       "      <td>0.656</td>\n",
       "      <td>0.688</td>\n",
       "      <td>0.371</td>\n",
       "      <td>0.388</td>\n",
       "      <td>0.518</td>\n",
       "      <td>0.495</td>\n",
       "      <td>0.4215</td>\n",
       "      <td>0.4285</td>\n",
       "      <td>0.274401</td>\n",
       "      <td>0.295939</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>6000</td>\n",
       "      <td>0.428390</td>\n",
       "      <td>0.319</td>\n",
       "      <td>0.311</td>\n",
       "      <td>0.372</td>\n",
       "      <td>0.431</td>\n",
       "      <td>0.202</td>\n",
       "      <td>0.352</td>\n",
       "      <td>0.660</td>\n",
       "      <td>0.670</td>\n",
       "      <td>0.373</td>\n",
       "      <td>0.392</td>\n",
       "      <td>0.520</td>\n",
       "      <td>0.519</td>\n",
       "      <td>0.4315</td>\n",
       "      <td>0.4460</td>\n",
       "      <td>0.288591</td>\n",
       "      <td>0.306123</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>8000</td>\n",
       "      <td>0.443615</td>\n",
       "      <td>0.340</td>\n",
       "      <td>0.311</td>\n",
       "      <td>0.379</td>\n",
       "      <td>0.463</td>\n",
       "      <td>0.204</td>\n",
       "      <td>0.360</td>\n",
       "      <td>0.681</td>\n",
       "      <td>0.700</td>\n",
       "      <td>0.384</td>\n",
       "      <td>0.404</td>\n",
       "      <td>0.517</td>\n",
       "      <td>0.517</td>\n",
       "      <td>0.4630</td>\n",
       "      <td>0.4790</td>\n",
       "      <td>0.299186</td>\n",
       "      <td>0.314921</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>10000</td>\n",
       "      <td>0.441457</td>\n",
       "      <td>0.346</td>\n",
       "      <td>0.317</td>\n",
       "      <td>0.390</td>\n",
       "      <td>0.454</td>\n",
       "      <td>0.222</td>\n",
       "      <td>0.364</td>\n",
       "      <td>0.690</td>\n",
       "      <td>0.696</td>\n",
       "      <td>0.366</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.514</td>\n",
       "      <td>0.506</td>\n",
       "      <td>0.4890</td>\n",
       "      <td>0.4820</td>\n",
       "      <td>0.302189</td>\n",
       "      <td>0.317653</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       runname  steps  agg_score  commonsense_qa/acc  commonsense_qa/acc_norm  \\\n",
       "0  FineWeb-Edu   2000   0.390326               0.284                    0.283   \n",
       "1  FineWeb-Edu   4000   0.414680               0.322                    0.307   \n",
       "2  FineWeb-Edu   6000   0.428390               0.319                    0.311   \n",
       "3  FineWeb-Edu   8000   0.443615               0.340                    0.311   \n",
       "4  FineWeb-Edu  10000   0.441457               0.346                    0.317   \n",
       "\n",
       "   hellaswag/acc  hellaswag/acc_norm  openbookqa/acc  openbookqa/acc_norm  \\\n",
       "0          0.314               0.325           0.164                0.296   \n",
       "1          0.343               0.395           0.196                0.320   \n",
       "2          0.372               0.431           0.202                0.352   \n",
       "3          0.379               0.463           0.204                0.360   \n",
       "4          0.390               0.454           0.222                0.364   \n",
       "\n",
       "   piqa/acc  piqa/acc_norm  siqa/acc  siqa/acc_norm  winogrande/acc  \\\n",
       "0     0.623          0.632     0.362          0.406           0.511   \n",
       "1     0.656          0.688     0.371          0.388           0.518   \n",
       "2     0.660          0.670     0.373          0.392           0.520   \n",
       "3     0.681          0.700     0.384          0.404           0.517   \n",
       "4     0.690          0.696     0.366          0.395           0.514   \n",
       "\n",
       "   winogrande/acc_norm  arc/acc  arc/acc_norm  mmlu/acc  mmlu/acc_norm  \n",
       "0                0.511   0.3795        0.3850  0.265997       0.284605  \n",
       "1                0.495   0.4215        0.4285  0.274401       0.295939  \n",
       "2                0.519   0.4315        0.4460  0.288591       0.306123  \n",
       "3                0.517   0.4630        0.4790  0.299186       0.314921  \n",
       "4                0.506   0.4890        0.4820  0.302189       0.317653  "
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "# load guilherme csv with all the FW runs\n",
    "df = pd.read_csv(\"../src_data/eval_results.csv\")\n",
    "\n",
    "# load FineWeb-Edu csv\n",
    "df_2 = pd.read_csv(\"./loubna-edu_fw_ablations_metrics.csv\")\n",
    "df_2['runname'] = df_2['runname'].replace('edu_fineweb_350b_tokens-seed-1', 'FineWeb-Edu', regex=True)\n",
    "df_2.drop([\"seed\", \"all/acc\", \"all/acc_norm\"], axis=1, inplace=True)\n",
    "df_2.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "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>runname</th>\n",
       "      <th>steps</th>\n",
       "      <th>agg_score</th>\n",
       "      <th>commonsense_qa/acc</th>\n",
       "      <th>commonsense_qa/acc_norm</th>\n",
       "      <th>hellaswag/acc</th>\n",
       "      <th>hellaswag/acc_norm</th>\n",
       "      <th>openbookqa/acc</th>\n",
       "      <th>openbookqa/acc_norm</th>\n",
       "      <th>piqa/acc</th>\n",
       "      <th>...</th>\n",
       "      <th>siqa/acc</th>\n",
       "      <th>siqa/acc_norm</th>\n",
       "      <th>winogrande/acc</th>\n",
       "      <th>winogrande/acc_norm</th>\n",
       "      <th>sciq/acc</th>\n",
       "      <th>sciq/acc_norm</th>\n",
       "      <th>arc/acc</th>\n",
       "      <th>arc/acc_norm</th>\n",
       "      <th>mmlu/acc</th>\n",
       "      <th>mmlu/acc_norm</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1253</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>160000</td>\n",
       "      <td>0.507129</td>\n",
       "      <td>0.430</td>\n",
       "      <td>0.359</td>\n",
       "      <td>0.473</td>\n",
       "      <td>0.593</td>\n",
       "      <td>0.282</td>\n",
       "      <td>0.418</td>\n",
       "      <td>0.744</td>\n",
       "      <td>...</td>\n",
       "      <td>0.392</td>\n",
       "      <td>0.402</td>\n",
       "      <td>0.576</td>\n",
       "      <td>0.575</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.5670</td>\n",
       "      <td>0.5725</td>\n",
       "      <td>0.350226</td>\n",
       "      <td>0.374533</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1254</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>162000</td>\n",
       "      <td>0.509118</td>\n",
       "      <td>0.416</td>\n",
       "      <td>0.367</td>\n",
       "      <td>0.474</td>\n",
       "      <td>0.592</td>\n",
       "      <td>0.288</td>\n",
       "      <td>0.408</td>\n",
       "      <td>0.747</td>\n",
       "      <td>...</td>\n",
       "      <td>0.390</td>\n",
       "      <td>0.409</td>\n",
       "      <td>0.572</td>\n",
       "      <td>0.577</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.5720</td>\n",
       "      <td>0.5780</td>\n",
       "      <td>0.348268</td>\n",
       "      <td>0.372947</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1255</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>164000</td>\n",
       "      <td>0.507843</td>\n",
       "      <td>0.416</td>\n",
       "      <td>0.365</td>\n",
       "      <td>0.467</td>\n",
       "      <td>0.591</td>\n",
       "      <td>0.276</td>\n",
       "      <td>0.408</td>\n",
       "      <td>0.737</td>\n",
       "      <td>...</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.406</td>\n",
       "      <td>0.576</td>\n",
       "      <td>0.580</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.5635</td>\n",
       "      <td>0.5715</td>\n",
       "      <td>0.349943</td>\n",
       "      <td>0.372246</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1256</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>166000</td>\n",
       "      <td>0.508308</td>\n",
       "      <td>0.415</td>\n",
       "      <td>0.364</td>\n",
       "      <td>0.472</td>\n",
       "      <td>0.593</td>\n",
       "      <td>0.282</td>\n",
       "      <td>0.414</td>\n",
       "      <td>0.740</td>\n",
       "      <td>...</td>\n",
       "      <td>0.401</td>\n",
       "      <td>0.408</td>\n",
       "      <td>0.575</td>\n",
       "      <td>0.570</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.5640</td>\n",
       "      <td>0.5760</td>\n",
       "      <td>0.352203</td>\n",
       "      <td>0.373463</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1257</th>\n",
       "      <td>FineWeb-Edu</td>\n",
       "      <td>167000</td>\n",
       "      <td>0.509494</td>\n",
       "      <td>0.429</td>\n",
       "      <td>0.362</td>\n",
       "      <td>0.472</td>\n",
       "      <td>0.597</td>\n",
       "      <td>0.290</td>\n",
       "      <td>0.418</td>\n",
       "      <td>0.738</td>\n",
       "      <td>...</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.404</td>\n",
       "      <td>0.582</td>\n",
       "      <td>0.578</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.5670</td>\n",
       "      <td>0.5735</td>\n",
       "      <td>0.350671</td>\n",
       "      <td>0.374453</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          runname   steps  agg_score  commonsense_qa/acc  \\\n",
       "1253  FineWeb-Edu  160000   0.507129               0.430   \n",
       "1254  FineWeb-Edu  162000   0.509118               0.416   \n",
       "1255  FineWeb-Edu  164000   0.507843               0.416   \n",
       "1256  FineWeb-Edu  166000   0.508308               0.415   \n",
       "1257  FineWeb-Edu  167000   0.509494               0.429   \n",
       "\n",
       "      commonsense_qa/acc_norm  hellaswag/acc  hellaswag/acc_norm  \\\n",
       "1253                    0.359          0.473               0.593   \n",
       "1254                    0.367          0.474               0.592   \n",
       "1255                    0.365          0.467               0.591   \n",
       "1256                    0.364          0.472               0.593   \n",
       "1257                    0.362          0.472               0.597   \n",
       "\n",
       "      openbookqa/acc  openbookqa/acc_norm  piqa/acc  ...  siqa/acc  \\\n",
       "1253           0.282                0.418     0.744  ...     0.392   \n",
       "1254           0.288                0.408     0.747  ...     0.390   \n",
       "1255           0.276                0.408     0.737  ...     0.395   \n",
       "1256           0.282                0.414     0.740  ...     0.401   \n",
       "1257           0.290                0.418     0.738  ...     0.395   \n",
       "\n",
       "      siqa/acc_norm  winogrande/acc  winogrande/acc_norm  sciq/acc  \\\n",
       "1253          0.402           0.576                0.575       NaN   \n",
       "1254          0.409           0.572                0.577       NaN   \n",
       "1255          0.406           0.576                0.580       NaN   \n",
       "1256          0.408           0.575                0.570       NaN   \n",
       "1257          0.404           0.582                0.578       NaN   \n",
       "\n",
       "      sciq/acc_norm  arc/acc  arc/acc_norm  mmlu/acc  mmlu/acc_norm  \n",
       "1253            NaN   0.5670        0.5725  0.350226       0.374533  \n",
       "1254            NaN   0.5720        0.5780  0.348268       0.372947  \n",
       "1255            NaN   0.5635        0.5715  0.349943       0.372246  \n",
       "1256            NaN   0.5640        0.5760  0.352203       0.373463  \n",
       "1257            NaN   0.5670        0.5735  0.350671       0.374453  \n",
       "\n",
       "[5 rows x 21 columns]"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_full = pd.concat([df, df_2], ignore_index=True)\n",
    "df_full.tail()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Guilherme-Board plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "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>steps</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>runname</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>C4</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Dolma</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FineWeb</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FineWeb-Edu</th>\n",
       "      <td>82</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>RedPajama2</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>RefinedWeb</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SlimPajama</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>The Pile</th>\n",
       "      <td>168</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             steps\n",
       "runname           \n",
       "C4             168\n",
       "Dolma          168\n",
       "FineWeb        168\n",
       "FineWeb-Edu     82\n",
       "RedPajama2     168\n",
       "RefinedWeb     168\n",
       "SlimPajama     168\n",
       "The Pile       168"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_full.groupby(\"runname\").agg({\"steps\": \"count\"})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "fineweb_edu_steps = df_full[df_full[\"runname\"] == \"FineWeb-Edu\"][\"steps\"].unique()\n",
    "# Only selects steps that are in the fineweb_edu_steps  \n",
    "df_full = df_full[df_full[\"steps\"].isin(fineweb_edu_steps)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import json\n",
    "from matplotlib import pyplot as plt\n",
    "metrics = ['agg_score', 'commonsense_qa/acc_norm', 'hellaswag/acc_norm', 'openbookqa/acc_norm', 'piqa/acc_norm',\n",
    "                   'siqa/acc_norm', 'winogrande/acc_norm', 'arc/acc_norm', 'mmlu/acc_norm']\n",
    "\n",
    "def normalize_runname(runname):\n",
    "    return runname.replace(\"/\", \"_\")\n",
    "\n",
    "grouped = (\n",
    "    df_full.groupby([\"runname\", \"steps\"])\n",
    "    .agg(\n",
    "        {\n",
    "            key: \"mean\" for key in metrics\n",
    "        }\n",
    "    )\n",
    "    .reset_index()\n",
    ")\n",
    "\n",
    "file_id=\"../assets/data/plots/edu_ablations\"\n",
    "files = {}\n",
    "for metric in metrics:\n",
    "    datas = {}\n",
    "    for name, group in grouped.groupby(\"runname\"):\n",
    "        group = group[[\"steps\", metric]].sort_values(by=\"steps\")\n",
    "        group = group.set_index(\"steps\")\n",
    "        rolling_avg = group\n",
    "        # rolling_avg = group.rolling(window=5).mean()\n",
    "        datas[name] = {\n",
    "            \"x\": (rolling_avg.index * 2048 * 1024 * 1e-9).tolist(),\n",
    "            \"y\": rolling_avg[metric].tolist(),\n",
    "            \"label\": name,\n",
    "        }\n",
    "    # Sort the datata based on the steps\n",
    "    datas = {k: v for k, v in sorted(datas.items(), key=lambda x: -x[1][\"y\"][-1])}\n",
    "    # Create a folder\n",
    "    os.makedirs(f\"{file_id}\", exist_ok=True)\n",
    "    with open(f\"{file_id}/{normalize_runname(metric)}.json\", \"w\") as f:\n",
    "        json.dump({\n",
    "            \"data\": datas,\n",
    "            \"layout\": {\n",
    "                \"title\": {\n",
    "                    \"text\": \"Dataset ablations\"\n",
    "                },\n",
    "            }\n",
    "        }, f)\n",
    "    files[metric] = {\"file\": f\"{normalize_runname(metric)}.json\"}\n",
    "# Create index\n",
    "with open(f\"{file_id}/index.json\", \"w\") as f:\n",
    "    json.dump({\n",
    "        \"files\": files,\n",
    "        \"settings\": {\n",
    "            \"defaultMetric\": \"agg_score\",\n",
    "            \"slider\":{\"min\":0,\"max\":30,\"default\":5},\n",
    "            \"caption\": \"📚 FineWeb-Edu outperforms 🍷 FineWeb and all other open web datasets on our group of evaluation tasks.\"\n",
    "        }\n",
    "    }, f)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Barplot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: kaleido in /Users/hynky/.pyenv/versions/3.12.2/envs/datatrove/lib/python3.12/site-packages (0.2.1)\n"
     ]
    }
   ],
   "source": [
    "!pip install -U kaleido"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot saved to plots/edu-100k.png\n"
     ]
    }
   ],
   "source": [
    "import plotly.express as px\n",
    "from plotly.subplots import make_subplots\n",
    "import plotly.graph_objects as go\n",
    "\n",
    "import json\n",
    "\n",
    "BASELINES = {\n",
    "    \"mmlu/acc_norm\": 0.25,\n",
    "    \"arc/acc_norm\": 0.25,\n",
    "    \"openbookqa/acc_norm\": 0.25,\n",
    "    \"piqa/acc_norm\": 0.5,\n",
    "    \"hellaswag/acc_norm\": 0.25,\n",
    "    \"siqa/acc_norm\": 0.33,\n",
    "    \"winogrande/acc_norm\": 0.5,\n",
    "}\n",
    "\n",
    "\n",
    "def normalize_run_name(run_name):\n",
    "    return run_name.replace(\"/\", \"_\")\n",
    "\n",
    "\n",
    "def save_for_bar(dir_name, df, metrics, default_metric=\"mmlu/acc_norm\", xlabel=\"Dataset\", plot_name=\"plot name\", custom_layout={}, ranges={}):\n",
    "    import os\n",
    "    files = {}\n",
    "    os.makedirs(f\"../assets/data/plots/{dir_name}\", exist_ok=True)\n",
    "    for metric in metrics:\n",
    "        data = {}\n",
    "        for run_name in df[\"runname\"].unique():\n",
    "            data[run_name] = {\n",
    "                \"x\": [run_name],\n",
    "                \"y\": df[df[\"runname\"] == run_name][metric].tolist(),\n",
    "                \"label\": run_name,\n",
    "            }\n",
    "        file_name = f\"{normalize_run_name(metric)}.json\"\n",
    "        files[metric] = {\"file\": f\"{file_name}\"}\n",
    "        with open(f\"../assets/data/plots/{dir_name}/{file_name}\", \"w\") as f:\n",
    "            json.dump({\n",
    "                \"data\": data,\n",
    "                \"layout\": {\n",
    "                    \"showlegend\": False,\n",
    "                    \"title\": {\n",
    "                        \"text\": plot_name,\n",
    "                    },\n",
    "                    \"xaxis\": {\n",
    "                        \"title\": {\n",
    "                            \"text\": xlabel,\n",
    "                            \"standoff\": 30\n",
    "                        },\n",
    "                        \"tickangle\": 30\n",
    "                    },\n",
    "                    \"yaxis\": {\n",
    "                        \"range\": ranges.get(metric, [0, 1])\n",
    "                    },\n",
    "                    \"margin\": {\n",
    "                        \"b\": 100\n",
    "                    },\n",
    "                    **custom_layout,\n",
    "                }\n",
    "            }, f)\n",
    "    with open(f\"../assets/data/plots/{dir_name}/index.json\", \"w\") as f:\n",
    "        json.dump({\n",
    "            \"files\": files,\n",
    "            \"settings\": {\n",
    "                \"defaultMetric\": default_metric,\n",
    "                \"slider\": None,\n",
    "                \"autoSetXRange\": False,\n",
    "                \"type\": \"bar\"\n",
    "            }\n",
    "        }, f)\n",
    "    return files\n",
    "\n",
    "def plot_metric_comparison(df, step, metrics, plot_name, run_name_replacements=None, output_file='comparison_plot_percentages.png', default_metric=\"mmlu/acc_norm\", custom_layout={}):\n",
    "    \"\"\"\n",
    "    Plot a comparison of the given metrics across different runs at the specified step and save the plot.\n",
    "    \"\"\"\n",
    "    if run_name_replacements:\n",
    "        df['runname'] = df['runname'].replace(run_name_replacements)\n",
    "\n",
    "    df_filtered = df[df['steps'] == step]\n",
    "\n",
    "    # Create subplots\n",
    "\n",
    "\n",
    "    ranges = {}\n",
    "    for i, metric in enumerate(metrics):\n",
    "        yrange_start = BASELINES.get(metric, 0) * 0.9\n",
    "        yrange_end = max(df_filtered[metric])\n",
    "        # Adjust the end\n",
    "        yrange_end = yrange_end + (yrange_end - yrange_start) * 0.2\n",
    "        ranges[metric] = [yrange_start, yrange_end]\n",
    "    \n",
    "    file_name=f\"plots/{output_file}.png\"\n",
    "    # fig.write_image(file_name)\n",
    "    print(f\"Plot saved to {file_name}\")\n",
    "\n",
    "    save_for_bar(output_file, df_filtered, metrics, default_metric, plot_name=plot_name, custom_layout=custom_layout, ranges=ranges)\n",
    "\n",
    "\n",
    "metrics = [\n",
    "    \"mmlu/acc_norm\",\n",
    "    \"arc/acc_norm\",\n",
    "    \"openbookqa/acc_norm\",\n",
    "    \"piqa/acc_norm\",\n",
    "    \"hellaswag/acc_norm\",\n",
    "    \"siqa/acc_norm\",\n",
    "    \"winogrande/acc_norm\",\n",
    "]\n",
    "\n",
    "plot_metric_comparison(df_full, 167000, metrics, output_file=\"edu-100k\", plot_name=\"Evaluation results at 350B tokens\", run_name_replacements={\n",
    "    \"FineWeb (ours)\": \"FineWeb\"\n",
    "})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Thresholds ablation"
   ]
  },
  {
   "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>runname</th>\n",
       "      <th>steps</th>\n",
       "      <th>agg_score</th>\n",
       "      <th>commonsense_qa/acc</th>\n",
       "      <th>commonsense_qa/acc_norm</th>\n",
       "      <th>hellaswag/acc</th>\n",
       "      <th>hellaswag/acc_norm</th>\n",
       "      <th>openbookqa/acc</th>\n",
       "      <th>openbookqa/acc_norm</th>\n",
       "      <th>piqa/acc</th>\n",
       "      <th>...</th>\n",
       "      <th>siqa/acc</th>\n",
       "      <th>siqa/acc_norm</th>\n",
       "      <th>winogrande/acc</th>\n",
       "      <th>winogrande/acc_norm</th>\n",
       "      <th>sciq/acc</th>\n",
       "      <th>sciq/acc_norm</th>\n",
       "      <th>arc/acc</th>\n",
       "      <th>arc/acc_norm</th>\n",
       "      <th>mmlu/acc</th>\n",
       "      <th>mmlu/acc_norm</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>C4</td>\n",
       "      <td>0</td>\n",
       "      <td>0.330893</td>\n",
       "      <td>0.186</td>\n",
       "      <td>0.233</td>\n",
       "      <td>0.272</td>\n",
       "      <td>0.258</td>\n",
       "      <td>0.166</td>\n",
       "      <td>0.286</td>\n",
       "      <td>0.542</td>\n",
       "      <td>...</td>\n",
       "      <td>0.367</td>\n",
       "      <td>0.362</td>\n",
       "      <td>0.516</td>\n",
       "      <td>0.497</td>\n",
       "      <td>0.208</td>\n",
       "      <td>0.202</td>\n",
       "      <td>0.2195</td>\n",
       "      <td>0.2510</td>\n",
       "      <td>0.230294</td>\n",
       "      <td>0.250147</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>C4</td>\n",
       "      <td>1000</td>\n",
       "      <td>0.355112</td>\n",
       "      <td>0.229</td>\n",
       "      <td>0.260</td>\n",
       "      <td>0.286</td>\n",
       "      <td>0.288</td>\n",
       "      <td>0.128</td>\n",
       "      <td>0.250</td>\n",
       "      <td>0.614</td>\n",
       "      <td>...</td>\n",
       "      <td>0.351</td>\n",
       "      <td>0.404</td>\n",
       "      <td>0.519</td>\n",
       "      <td>0.476</td>\n",
       "      <td>0.565</td>\n",
       "      <td>0.518</td>\n",
       "      <td>0.2680</td>\n",
       "      <td>0.2935</td>\n",
       "      <td>0.238951</td>\n",
       "      <td>0.250399</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>C4</td>\n",
       "      <td>2000</td>\n",
       "      <td>0.378435</td>\n",
       "      <td>0.268</td>\n",
       "      <td>0.278</td>\n",
       "      <td>0.312</td>\n",
       "      <td>0.330</td>\n",
       "      <td>0.122</td>\n",
       "      <td>0.276</td>\n",
       "      <td>0.646</td>\n",
       "      <td>...</td>\n",
       "      <td>0.375</td>\n",
       "      <td>0.400</td>\n",
       "      <td>0.509</td>\n",
       "      <td>0.500</td>\n",
       "      <td>0.676</td>\n",
       "      <td>0.577</td>\n",
       "      <td>0.3065</td>\n",
       "      <td>0.3230</td>\n",
       "      <td>0.247275</td>\n",
       "      <td>0.255482</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>C4</td>\n",
       "      <td>3000</td>\n",
       "      <td>0.387795</td>\n",
       "      <td>0.280</td>\n",
       "      <td>0.295</td>\n",
       "      <td>0.331</td>\n",
       "      <td>0.380</td>\n",
       "      <td>0.152</td>\n",
       "      <td>0.274</td>\n",
       "      <td>0.660</td>\n",
       "      <td>...</td>\n",
       "      <td>0.376</td>\n",
       "      <td>0.387</td>\n",
       "      <td>0.512</td>\n",
       "      <td>0.496</td>\n",
       "      <td>0.725</td>\n",
       "      <td>0.621</td>\n",
       "      <td>0.3175</td>\n",
       "      <td>0.3340</td>\n",
       "      <td>0.254534</td>\n",
       "      <td>0.267363</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>C4</td>\n",
       "      <td>4000</td>\n",
       "      <td>0.399320</td>\n",
       "      <td>0.296</td>\n",
       "      <td>0.298</td>\n",
       "      <td>0.351</td>\n",
       "      <td>0.406</td>\n",
       "      <td>0.168</td>\n",
       "      <td>0.282</td>\n",
       "      <td>0.676</td>\n",
       "      <td>...</td>\n",
       "      <td>0.382</td>\n",
       "      <td>0.404</td>\n",
       "      <td>0.522</td>\n",
       "      <td>0.503</td>\n",
       "      <td>0.723</td>\n",
       "      <td>0.618</td>\n",
       "      <td>0.3255</td>\n",
       "      <td>0.3470</td>\n",
       "      <td>0.254762</td>\n",
       "      <td>0.263563</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",
       "      <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",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1171</th>\n",
       "      <td>The Pile</td>\n",
       "      <td>163000</td>\n",
       "      <td>0.463789</td>\n",
       "      <td>0.379</td>\n",
       "      <td>0.349</td>\n",
       "      <td>0.441</td>\n",
       "      <td>0.555</td>\n",
       "      <td>0.240</td>\n",
       "      <td>0.366</td>\n",
       "      <td>0.701</td>\n",
       "      <td>...</td>\n",
       "      <td>0.405</td>\n",
       "      <td>0.388</td>\n",
       "      <td>0.585</td>\n",
       "      <td>0.560</td>\n",
       "      <td>0.875</td>\n",
       "      <td>0.820</td>\n",
       "      <td>0.4475</td>\n",
       "      <td>0.4450</td>\n",
       "      <td>0.299378</td>\n",
       "      <td>0.326313</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1172</th>\n",
       "      <td>The Pile</td>\n",
       "      <td>164000</td>\n",
       "      <td>0.462758</td>\n",
       "      <td>0.369</td>\n",
       "      <td>0.344</td>\n",
       "      <td>0.438</td>\n",
       "      <td>0.552</td>\n",
       "      <td>0.248</td>\n",
       "      <td>0.348</td>\n",
       "      <td>0.708</td>\n",
       "      <td>...</td>\n",
       "      <td>0.395</td>\n",
       "      <td>0.401</td>\n",
       "      <td>0.577</td>\n",
       "      <td>0.567</td>\n",
       "      <td>0.874</td>\n",
       "      <td>0.806</td>\n",
       "      <td>0.4465</td>\n",
       "      <td>0.4355</td>\n",
       "      <td>0.302083</td>\n",
       "      <td>0.331563</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1173</th>\n",
       "      <td>The Pile</td>\n",
       "      <td>165000</td>\n",
       "      <td>0.465026</td>\n",
       "      <td>0.383</td>\n",
       "      <td>0.350</td>\n",
       "      <td>0.438</td>\n",
       "      <td>0.553</td>\n",
       "      <td>0.234</td>\n",
       "      <td>0.352</td>\n",
       "      <td>0.707</td>\n",
       "      <td>...</td>\n",
       "      <td>0.400</td>\n",
       "      <td>0.401</td>\n",
       "      <td>0.569</td>\n",
       "      <td>0.556</td>\n",
       "      <td>0.874</td>\n",
       "      <td>0.811</td>\n",
       "      <td>0.4460</td>\n",
       "      <td>0.4455</td>\n",
       "      <td>0.305193</td>\n",
       "      <td>0.331708</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1174</th>\n",
       "      <td>The Pile</td>\n",
       "      <td>166000</td>\n",
       "      <td>0.462349</td>\n",
       "      <td>0.377</td>\n",
       "      <td>0.346</td>\n",
       "      <td>0.440</td>\n",
       "      <td>0.557</td>\n",
       "      <td>0.228</td>\n",
       "      <td>0.346</td>\n",
       "      <td>0.711</td>\n",
       "      <td>...</td>\n",
       "      <td>0.398</td>\n",
       "      <td>0.398</td>\n",
       "      <td>0.572</td>\n",
       "      <td>0.558</td>\n",
       "      <td>0.877</td>\n",
       "      <td>0.811</td>\n",
       "      <td>0.4525</td>\n",
       "      <td>0.4385</td>\n",
       "      <td>0.301952</td>\n",
       "      <td>0.331295</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1175</th>\n",
       "      <td>The Pile</td>\n",
       "      <td>167000</td>\n",
       "      <td>0.464539</td>\n",
       "      <td>0.386</td>\n",
       "      <td>0.354</td>\n",
       "      <td>0.434</td>\n",
       "      <td>0.557</td>\n",
       "      <td>0.232</td>\n",
       "      <td>0.356</td>\n",
       "      <td>0.706</td>\n",
       "      <td>...</td>\n",
       "      <td>0.402</td>\n",
       "      <td>0.402</td>\n",
       "      <td>0.573</td>\n",
       "      <td>0.559</td>\n",
       "      <td>0.867</td>\n",
       "      <td>0.802</td>\n",
       "      <td>0.4475</td>\n",
       "      <td>0.4375</td>\n",
       "      <td>0.301934</td>\n",
       "      <td>0.330810</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1176 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       runname   steps  agg_score  commonsense_qa/acc  \\\n",
       "0           C4       0   0.330893               0.186   \n",
       "1           C4    1000   0.355112               0.229   \n",
       "2           C4    2000   0.378435               0.268   \n",
       "3           C4    3000   0.387795               0.280   \n",
       "4           C4    4000   0.399320               0.296   \n",
       "...        ...     ...        ...                 ...   \n",
       "1171  The Pile  163000   0.463789               0.379   \n",
       "1172  The Pile  164000   0.462758               0.369   \n",
       "1173  The Pile  165000   0.465026               0.383   \n",
       "1174  The Pile  166000   0.462349               0.377   \n",
       "1175  The Pile  167000   0.464539               0.386   \n",
       "\n",
       "      commonsense_qa/acc_norm  hellaswag/acc  hellaswag/acc_norm  \\\n",
       "0                       0.233          0.272               0.258   \n",
       "1                       0.260          0.286               0.288   \n",
       "2                       0.278          0.312               0.330   \n",
       "3                       0.295          0.331               0.380   \n",
       "4                       0.298          0.351               0.406   \n",
       "...                       ...            ...                 ...   \n",
       "1171                    0.349          0.441               0.555   \n",
       "1172                    0.344          0.438               0.552   \n",
       "1173                    0.350          0.438               0.553   \n",
       "1174                    0.346          0.440               0.557   \n",
       "1175                    0.354          0.434               0.557   \n",
       "\n",
       "      openbookqa/acc  openbookqa/acc_norm  piqa/acc  ...  siqa/acc  \\\n",
       "0              0.166                0.286     0.542  ...     0.367   \n",
       "1              0.128                0.250     0.614  ...     0.351   \n",
       "2              0.122                0.276     0.646  ...     0.375   \n",
       "3              0.152                0.274     0.660  ...     0.376   \n",
       "4              0.168                0.282     0.676  ...     0.382   \n",
       "...              ...                  ...       ...  ...       ...   \n",
       "1171           0.240                0.366     0.701  ...     0.405   \n",
       "1172           0.248                0.348     0.708  ...     0.395   \n",
       "1173           0.234                0.352     0.707  ...     0.400   \n",
       "1174           0.228                0.346     0.711  ...     0.398   \n",
       "1175           0.232                0.356     0.706  ...     0.402   \n",
       "\n",
       "      siqa/acc_norm  winogrande/acc  winogrande/acc_norm  sciq/acc  \\\n",
       "0             0.362           0.516                0.497     0.208   \n",
       "1             0.404           0.519                0.476     0.565   \n",
       "2             0.400           0.509                0.500     0.676   \n",
       "3             0.387           0.512                0.496     0.725   \n",
       "4             0.404           0.522                0.503     0.723   \n",
       "...             ...             ...                  ...       ...   \n",
       "1171          0.388           0.585                0.560     0.875   \n",
       "1172          0.401           0.577                0.567     0.874   \n",
       "1173          0.401           0.569                0.556     0.874   \n",
       "1174          0.398           0.572                0.558     0.877   \n",
       "1175          0.402           0.573                0.559     0.867   \n",
       "\n",
       "      sciq/acc_norm  arc/acc  arc/acc_norm  mmlu/acc  mmlu/acc_norm  \n",
       "0             0.202   0.2195        0.2510  0.230294       0.250147  \n",
       "1             0.518   0.2680        0.2935  0.238951       0.250399  \n",
       "2             0.577   0.3065        0.3230  0.247275       0.255482  \n",
       "3             0.621   0.3175        0.3340  0.254534       0.267363  \n",
       "4             0.618   0.3255        0.3470  0.254762       0.263563  \n",
       "...             ...      ...           ...       ...            ...  \n",
       "1171          0.820   0.4475        0.4450  0.299378       0.326313  \n",
       "1172          0.806   0.4465        0.4355  0.302083       0.331563  \n",
       "1173          0.811   0.4460        0.4455  0.305193       0.331708  \n",
       "1174          0.811   0.4525        0.4385  0.301952       0.331295  \n",
       "1175          0.802   0.4475        0.4375  0.301934       0.330810  \n",
       "\n",
       "[1176 rows x 21 columns]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Fetching datafiles...: 100%|██████████| 4/4 [00:00<00:00, 21.68it/s]\n",
      "Loading evals data...: 100%|██████████| 26/26 [00:04<00:00,  5.76it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Metrics saved to loubna-ablations_faq_metrics.csv\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "token = os.getenv(\"HF_TOKEN\")\n",
    "repo_name = \"HuggingFaceTB/loubna-ablations_faq\"\n",
    "runs_to_fetch = [\"filtered_web_min_score_4_fix-seed-1-\", \"fineweb_2B_educational_minimum_score_3-seed-0-\", \"fineweb_2B_educational_regression-seed-6-\", \"fineweb_2024_10_all_2B-seed-6-\"]\n",
    "steps_to_fetch = \"%1000\"\n",
    "prefix = \"tb/ablations_faq-1p81G-\"\n",
    "metrics = ['commonsense_qa/acc_norm', 'hellaswag/acc_norm', 'openbookqa/acc_norm', 'piqa/acc_norm',\n",
    "                   'siqa/acc_norm', 'winogrande/acc_norm', 'arc/acc_norm', 'mmlu/acc_norm']\n",
    "agg_score_columns = metrics\n",
    "column_name = \"agg_score\"\n",
    "seed_merge_method = \"mean\"\n",
    "oauth_token = token\n",
    "\n",
    "runs_to_fetch = [prefix + run for run in runs_to_fetch]\n",
    "df = fetch_run_results_simple(repo_name, runs_to_fetch, steps_to_fetch, prefix, agg_score_columns, column_name, seed_merge_method, oauth_token=token)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>runname</th>\n",
       "      <th>seed</th>\n",
       "      <th>steps</th>\n",
       "      <th>agg_score</th>\n",
       "      <th>commonsense_qa/acc</th>\n",
       "      <th>commonsense_qa/acc_norm</th>\n",
       "      <th>hellaswag/acc</th>\n",
       "      <th>hellaswag/acc_norm</th>\n",
       "      <th>openbookqa/acc</th>\n",
       "      <th>openbookqa/acc_norm</th>\n",
       "      <th>...</th>\n",
       "      <th>siqa/acc</th>\n",
       "      <th>siqa/acc_norm</th>\n",
       "      <th>winogrande/acc</th>\n",
       "      <th>winogrande/acc_norm</th>\n",
       "      <th>all/acc</th>\n",
       "      <th>all/acc_norm</th>\n",
       "      <th>arc/acc</th>\n",
       "      <th>arc/acc_norm</th>\n",
       "      <th>mmlu/acc</th>\n",
       "      <th>mmlu/acc_norm</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>FineWeb (FW)</td>\n",
       "      <td>0</td>\n",
       "      <td>4000</td>\n",
       "      <td>0.389983</td>\n",
       "      <td>0.275</td>\n",
       "      <td>0.281</td>\n",
       "      <td>0.352</td>\n",
       "      <td>0.383</td>\n",
       "      <td>0.152</td>\n",
       "      <td>0.286</td>\n",
       "      <td>...</td>\n",
       "      <td>0.365</td>\n",
       "      <td>0.385</td>\n",
       "      <td>0.505</td>\n",
       "      <td>0.493</td>\n",
       "      <td>0.265054</td>\n",
       "      <td>0.281046</td>\n",
       "      <td>0.3265</td>\n",
       "      <td>0.3435</td>\n",
       "      <td>0.250500</td>\n",
       "      <td>0.264368</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>FineWeb (FW)</td>\n",
       "      <td>0</td>\n",
       "      <td>5000</td>\n",
       "      <td>0.397987</td>\n",
       "      <td>0.303</td>\n",
       "      <td>0.297</td>\n",
       "      <td>0.349</td>\n",
       "      <td>0.397</td>\n",
       "      <td>0.154</td>\n",
       "      <td>0.290</td>\n",
       "      <td>...</td>\n",
       "      <td>0.375</td>\n",
       "      <td>0.383</td>\n",
       "      <td>0.509</td>\n",
       "      <td>0.502</td>\n",
       "      <td>0.268548</td>\n",
       "      <td>0.282678</td>\n",
       "      <td>0.3340</td>\n",
       "      <td>0.3560</td>\n",
       "      <td>0.253134</td>\n",
       "      <td>0.264896</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>FineWeb (FW)</td>\n",
       "      <td>0</td>\n",
       "      <td>6000</td>\n",
       "      <td>0.403954</td>\n",
       "      <td>0.317</td>\n",
       "      <td>0.319</td>\n",
       "      <td>0.359</td>\n",
       "      <td>0.416</td>\n",
       "      <td>0.166</td>\n",
       "      <td>0.284</td>\n",
       "      <td>...</td>\n",
       "      <td>0.379</td>\n",
       "      <td>0.400</td>\n",
       "      <td>0.516</td>\n",
       "      <td>0.490</td>\n",
       "      <td>0.268197</td>\n",
       "      <td>0.286678</td>\n",
       "      <td>0.3330</td>\n",
       "      <td>0.3590</td>\n",
       "      <td>0.252102</td>\n",
       "      <td>0.268633</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>FineWeb (FW)</td>\n",
       "      <td>0</td>\n",
       "      <td>7000</td>\n",
       "      <td>0.404859</td>\n",
       "      <td>0.298</td>\n",
       "      <td>0.310</td>\n",
       "      <td>0.367</td>\n",
       "      <td>0.424</td>\n",
       "      <td>0.176</td>\n",
       "      <td>0.290</td>\n",
       "      <td>...</td>\n",
       "      <td>0.382</td>\n",
       "      <td>0.396</td>\n",
       "      <td>0.511</td>\n",
       "      <td>0.494</td>\n",
       "      <td>0.271701</td>\n",
       "      <td>0.289459</td>\n",
       "      <td>0.3250</td>\n",
       "      <td>0.3510</td>\n",
       "      <td>0.256203</td>\n",
       "      <td>0.271874</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>FineWeb (FW)</td>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>0.403283</td>\n",
       "      <td>0.330</td>\n",
       "      <td>0.319</td>\n",
       "      <td>0.364</td>\n",
       "      <td>0.412</td>\n",
       "      <td>0.176</td>\n",
       "      <td>0.276</td>\n",
       "      <td>...</td>\n",
       "      <td>0.383</td>\n",
       "      <td>0.403</td>\n",
       "      <td>0.510</td>\n",
       "      <td>0.493</td>\n",
       "      <td>0.267533</td>\n",
       "      <td>0.287018</td>\n",
       "      <td>0.3295</td>\n",
       "      <td>0.3510</td>\n",
       "      <td>0.251046</td>\n",
       "      <td>0.269266</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 22 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        runname  seed  steps  agg_score  commonsense_qa/acc  \\\n",
       "0  FineWeb (FW)     0   4000   0.389983               0.275   \n",
       "0  FineWeb (FW)     0   5000   0.397987               0.303   \n",
       "0  FineWeb (FW)     0   6000   0.403954               0.317   \n",
       "0  FineWeb (FW)     0   7000   0.404859               0.298   \n",
       "0  FineWeb (FW)     0   8000   0.403283               0.330   \n",
       "\n",
       "   commonsense_qa/acc_norm  hellaswag/acc  hellaswag/acc_norm  openbookqa/acc  \\\n",
       "0                    0.281          0.352               0.383           0.152   \n",
       "0                    0.297          0.349               0.397           0.154   \n",
       "0                    0.319          0.359               0.416           0.166   \n",
       "0                    0.310          0.367               0.424           0.176   \n",
       "0                    0.319          0.364               0.412           0.176   \n",
       "\n",
       "   openbookqa/acc_norm  ...  siqa/acc  siqa/acc_norm  winogrande/acc  \\\n",
       "0                0.286  ...     0.365          0.385           0.505   \n",
       "0                0.290  ...     0.375          0.383           0.509   \n",
       "0                0.284  ...     0.379          0.400           0.516   \n",
       "0                0.290  ...     0.382          0.396           0.511   \n",
       "0                0.276  ...     0.383          0.403           0.510   \n",
       "\n",
       "   winogrande/acc_norm   all/acc  all/acc_norm  arc/acc  arc/acc_norm  \\\n",
       "0                0.493  0.265054      0.281046   0.3265        0.3435   \n",
       "0                0.502  0.268548      0.282678   0.3340        0.3560   \n",
       "0                0.490  0.268197      0.286678   0.3330        0.3590   \n",
       "0                0.494  0.271701      0.289459   0.3250        0.3510   \n",
       "0                0.493  0.267533      0.287018   0.3295        0.3510   \n",
       "\n",
       "   mmlu/acc  mmlu/acc_norm  \n",
       "0  0.250500       0.264368  \n",
       "0  0.253134       0.264896  \n",
       "0  0.252102       0.268633  \n",
       "0  0.256203       0.271874  \n",
       "0  0.251046       0.269266  \n",
       "\n",
       "[5 rows x 22 columns]"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['runname'] = df['runname'].replace({\"filtered_web_min_score_4_fix-seed-1\": \"FW-Edu-threshold=4\",\n",
    "                                        \"fineweb_2B_educational_minimum_score_3-seed-0\": \"FW-Edu-threshold=3\",\n",
    "                                        \"fineweb_2B_educational_regression-seed-6\": \"FW-Edu-threshold=2\",\n",
    "                                        \"fineweb_2024_10_all_2B-seed-6\": \"FineWeb (FW)\"}, regex=True)\n",
    "df.tail()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=4\n",
       "0    FW-Edu-threshold=3\n",
       "0    FW-Edu-threshold=3\n",
       "0    FW-Edu-threshold=3\n",
       "0    FW-Edu-threshold=3\n",
       "0    FW-Edu-threshold=3\n",
       "0    FW-Edu-threshold=2\n",
       "0    FW-Edu-threshold=2\n",
       "0    FW-Edu-threshold=2\n",
       "0    FW-Edu-threshold=2\n",
       "0    FW-Edu-threshold=2\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "0          FineWeb (FW)\n",
       "Name: runname, dtype: object"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[\"runname\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot saved to plots/edu-8k.png\n"
     ]
    }
   ],
   "source": [
    "\n",
    "metrics = [\n",
    "    \"mmlu/acc_norm\",\n",
    "    \"arc/acc_norm\",\n",
    "    \"openbookqa/acc_norm\",\n",
    "    \"piqa/acc_norm\",\n",
    "    \"hellaswag/acc_norm\",\n",
    "    \"siqa/acc_norm\",\n",
    "    \"winogrande/acc_norm\",\n",
    "]\n",
    "plot_metric_comparison(df, 8000, metrics, output_file=\"edu-8k\", plot_name=\"FineWeb-Edu thresholding\", custom_layout={\n",
    "    \"xaxis\": {\n",
    "        \"title\": {\n",
    "            \"standoff\": 60,\n",
    "            \"text\": \"Dataset\"\n",
    "        },\n",
    "        \"tickangle\": 30\n",
    "    },\n",
    "    \"margin\": {\n",
    "        \"b\": 120\n",
    "    }\n",
    "})"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "textbooks",
   "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": "3.12.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}