Fixes from runpod

tools/notebook/gemma.ipynb

@@ -203,6 +203,17 @@
     "tokenizer = AutoTokenizer.from_pretrained(model_id, token=os.environ['SIA_HF_API_KEY'])"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4e91a67e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Save tokenizer first\n",
+    "tokenizer.save_pretrained(output_dir+\"_tokenizer\")"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -323,17 +334,6 @@
     "trainer.train()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "4e91a67e",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# Save tokenizer first\n",
-    "tokenizer.save_pretrained(output_dir+\"_tokenizer\")"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -408,6 +408,16 @@
     ")"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6930cf5d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!cp /root/models/gemma3_12b_tokenizer/* /root/models/gemma3_12b_merged/"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "2fdfdfba",
@@ -511,6 +521,7 @@
    "outputs": [],
    "source": [
     "%%bash\n",
+    "cd /usr/local/lib/\n",
     "llama-cli -m /root/models/gemma3_1b.gguf -p \"hi\""
    ]
   },
@@ -672,6 +683,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "%cd /usr/local/lib/\n",
+    "\n",
     "from llama_cpp import Llama\n",
     "\n",
     "# Load the model\n",