feat(experiments): update gpt3, claude, gemini results

walledai · Aug 3, 2024 · f0ae274 · f0ae274
1 parent ec9fc38
commit f0ae274
Showing 1 changed file with 29 additions and 4 deletions.
diff --git a/experiments/Aggregate Experimental Results.ipynb b/experiments/Aggregate Experimental Results.ipynb
@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": 2,
    "id": "882091db-451c-4bfc-8a93-ef26dde618b8",
    "metadata": {},
    "outputs": [],
@@ -12,7 +12,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 3,
    "id": "8a2ad73b-9d9f-4a24-bd1f-306dea2ccfe8",
    "metadata": {},
    "outputs": [
@@ -52,6 +52,27 @@
        "      <td>0.7</td>\n",
        "    </tr>\n",
        "    <tr>\n",
+       "      <th>chatgpt4</th>\n",
+       "      <td>0.975</td>\n",
+       "      <td>0.998077</td>\n",
+       "      <td>0.996364</td>\n",
+       "      <td>0.853333</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>claude</th>\n",
+       "      <td>1.0</td>\n",
+       "      <td>1.0</td>\n",
+       "      <td>1.0</td>\n",
+       "      <td>0.753333</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>gemini</th>\n",
+       "      <td>1.0</td>\n",
+       "      <td>1.0</td>\n",
+       "      <td>1.0</td>\n",
+       "      <td>0.644444</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
        "      <th>gemma-1.1-7b</th>\n",
        "      <td>0.965</td>\n",
        "      <td>0.994231</td>\n",
@@ -149,6 +170,9 @@
       "text/plain": [
        "                 harmbench  advbench     catqa    xstest\n",
        "aya-23               0.725  0.913462  0.898182       0.7\n",
+       "chatgpt4             0.975  0.998077  0.996364  0.853333\n",
+       "claude                 1.0       1.0       1.0  0.753333\n",
+       "gemini                 1.0       1.0       1.0  0.644444\n",
        "gemma-1.1-7b         0.965  0.994231  0.938182  0.626667\n",
        "gemma-7b              0.92  0.978846  0.961818      0.64\n",
        "gemma2-9b            0.995       1.0  0.994545       0.7\n",
@@ -164,7 +188,7 @@
        "qwen2-7b              0.94  0.998077  0.989091  0.853333"
       ]
      },
-     "execution_count": 5,
+     "execution_count": 3,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -180,7 +204,8 @@
     "        \"mistral-nemo-12b\", \"mistral-7b\", \"mixtral-8x7b\",\n",
     "        \"phi3-mini\", #\"phi3-small\",\n",
     "        \"qwen2-7b\", \"qwen2-1.5b\", \"qwen2-0.5b\",\n",
-    "        \"aya-23\"\n",
+    "        \"aya-23\",\n",
+    "        \"chatgpt4\", \"claude\", \"gemini\"\n",
     "        # \"yi-1.5-6b\"\n",
     "    ]:\n",
     "        try:\n",