Fix app

a.ipynb

@@ -0,0 +1,168 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "from pathlib import Path\n",
+    "\n",
+    "import gradio as gr\n",
+    "import pandas as pd"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 31,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def get_leaderboard_df():\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",
+    "        model_revision = \"_\".join(path_parts[1:4])\n",
+    "        task = path_parts[4].capitalize()\n",
+    "        # Extract timestamp from filepath\n",
+    "        timestamp = filepath.stem.split(\"_\")[-1][:-3]\n",
+    "        df.loc[model_revision, \"Timestamp\"] = timestamp\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",
+    "            # TruthfulQA has two metrics, so we need to pick the `mc2` one that's reported on the leaderboard\n",
+    "            if task == \"truthfulqa\":\n",
+    "                value = data[\"results\"][first_result_key][\"truthfulqa_mc2\"]\n",
+    "            else:\n",
+    "                first_metric_key = next(iter(data[\"results\"][first_result_key]))  # 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",
+    "            df.loc[model_revision, task] = value\n",
+    " \n",
+    "    df.insert(loc=0, column=\"Average\", value=df.mean(axis=1, numeric_only=True))\n",
+    "    df = df.sort_values(by=[\"Average\"], ascending=False)\n",
+    "    df = df.reset_index().rename(columns={\"index\": \"Model\"}).round(3)\n",
+    "    return df"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 32,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "df = get_leaderboard_df()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "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>Timestamp</th>\n",
+       "      <th>Average</th>\n",
+       "      <th>Truthfulqa</th>\n",
+       "      <th>Winogrande</th>\n",
+       "      <th>Gsm8k</th>\n",
+       "      <th>Hellaswag</th>\n",
+       "      <th>Arc</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>Qwen_Qwen1.5-0.5B-Chat_main</td>\n",
+       "      <td>2024-02-28T07-35-58.803</td>\n",
+       "      <td>0.296</td>\n",
+       "      <td>0.271</td>\n",
+       "      <td>0.519</td>\n",
+       "      <td>0.039</td>\n",
+       "      <td>0.363</td>\n",
+       "      <td>0.287</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "                         Model                Timestamp  Average  Truthfulqa  \\\n",
+       "0  Qwen_Qwen1.5-0.5B-Chat_main  2024-02-28T07-35-58.803    0.296       0.271   \n",
+       "\n",
+       "   Winogrande  Gsm8k  Hellaswag    Arc  \n",
+       "0       0.519  0.039      0.363  0.287  "
+      ]
+     },
+     "execution_count": 28,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df"
+   ]
+  },
+  {
+   "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
+}

app.py

@@ -7,7 +7,7 @@ import pandas as pd
 TITLE = """<h1 align="center" id="space-title">LLM Leaderboard for H4 Models</h1>"""
 
 DESCRIPTION = f"""
-Evaluation of H4 models across a diverse range of benchmarks from Eleuther's [LLM evaluation harness](https://github.com/EleutherAI/lm-evaluation-harness).
+Evaluation of H4 models across a diverse range of benchmarks from [LightEval](https://github.com/huggingface/lighteval)
 """
 
 
@@ -18,7 +18,7 @@ def get_leaderboard_df():
     models = set()
     for filepath in filepaths:
         path_parts = Path(filepath).parts
-        model_revision = "_".join(path_parts[1:4])  # Adjust indices based on your path structure
+        model_revision = "_".join(path_parts[1:4])
         models.add(model_revision)
 
     # Initialize DataFrame
@@ -27,17 +27,26 @@ def get_leaderboard_df():
     # Extract data from each file and populate the DataFrame
     for filepath in filepaths:
         path_parts = Path(filepath).parts
-        model_revision = "_".join(path_parts[1:4])  # Adjust indices based on your path structure
-        task = Path(filepath).stem.split("_")[-1].capitalize()  # gets 'mmlu' from the filename
+        model_revision = "_".join(path_parts[1:4])
+        task = path_parts[4].capitalize()
+        # Extract timestamp from filepath
+        timestamp = filepath.stem.split("_")[-1][:-3]
+        df.loc[model_revision, "Timestamp"] = timestamp
 
         with open(filepath, "r") as file:
             data = json.load(file)
             first_result_key = next(iter(data["results"]))  # gets the first key in 'results'
-            first_metric_key = next(iter(data["results"][first_result_key]))  # gets the first key in the first result
-            value = data["results"][first_result_key][first_metric_key]  # gets the value of the first metric
+            # TruthfulQA has two metrics, so we need to pick the `mc2` one that's reported on the leaderboard
+            if task == "truthfulqa":
+                value = data["results"][first_result_key]["truthfulqa_mc2"]
+            else:
+                first_metric_key = next(
+                    iter(data["results"][first_result_key])
+                )  # gets the first key in the first result
+                value = data["results"][first_result_key][first_metric_key]  # gets the value of the first metric
             df.loc[model_revision, task] = value
 
-    df.insert(loc=0, column="Average", value=df.mean(axis=1))
+    df.insert(loc=1, column="Average", value=df.mean(axis=1, numeric_only=True))
     df = df.sort_values(by=["Average"], ascending=False)
     df = df.reset_index().rename(columns={"index": "Model"}).round(3)
     return df