diff --git a/spkatt-2023.ipynb b/spkatt-2023.ipynb
index 2b3ff00..968a8e8 100644
--- a/spkatt-2023.ipynb
+++ b/spkatt-2023.ipynb
@@ -44,6 +44,8 @@
     "from tqdm import tqdm\n",
     "import json\n",
     "import warnings\n",
+    "import gc\n",
+    "import torch\n",
     "\n",
     "from datasets import (\n",
     "    load_dataset,\n",
@@ -1647,6 +1649,42 @@
     "- `save_steps` and `max_steps`: set `max_steps` to control the length of training (`save_steps` determines when checkpoints are created)\n"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# choose config for cue model\n",
+    "cues_training_config = \"./configs/7b_cues.args\"  # 7b model\n",
+    "# cues_training_config = \"./configs/70b_cues.args\" # 70b model\n",
+    "\n",
+    "train(cues_training_config)\n",
+    "\n",
+    "# free vram after training\n",
+    "gc.collect()\n",
+    "torch.cuda.empty_cache()\n",
+    "gc.collect()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# choose config for roles model\n",
+    "roles_training_config = \"./configs/7b_roles.args\"  # 7b model\n",
+    "# roles_training_config = \"./configs/70b_roles.args\" # 70b model\n",
+    "\n",
+    "train(roles_training_config)\n",
+    "\n",
+    "# free vram after training\n",
+    "gc.collect()\n",
+    "torch.cuda.empty_cache()\n",
+    "gc.collect()"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,