Modify lm similarity example notebook to work without cuda

Author: WarmCyan

notebooks/lm_similarity_example.ipynb

@@ -7,7 +7,22 @@
    "source": [
     "# Language Model Similarity Example\n",
     "\n",
-    "This notebook shows how to provide a language model to a similarity anchor, allowing the utilization of knowledge inside embedding spaces as part of the ICAT model."
+    "This notebook shows how to provide a language model to a similarity anchor, allowing the utilization of knowledge inside embedding spaces as part of the ICAT model.\n",
+    "\n",
+    "You will need to install the huggingface transformers and pytorch libraries for this notebook to run, please use\n",
+    "```\n",
+    "pip install transformers torch\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "bb6a33c6-e0f2-414f-9356-97f6fb47e2b9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch"
    ]
   },
   {
@@ -21,7 +36,7 @@
    "source": [
     "# change these constants as needed based on your hardware constraints\n",
     "BATCH_SIZE = 16\n",
-    "DEVICE = \"cuda\"\n",
+    "DEVICE = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
     "MODEL_NAME = \"bert-base-uncased\""
    ]
   },
@@ -175,7 +190,7 @@
     "\n",
     "dataset = fetch_20newsgroups(subset=\"train\")\n",
     "df = pd.DataFrame({\"text\": dataset[\"data\"], \"category\": [dataset[\"target_names\"][i] for i in dataset[\"target\"]]})\n",
-    "#df = df.iloc[0:1999]\n",
+    "df = df.iloc[0:1999]  # NOTE: if running on CPU or weaker GPU, recommend uncommenting this to avoid long processing times on first BERT anchor creation.\n",
     "df.head()"
    ]
   },
@@ -196,7 +211,7 @@
    },
    "outputs": [],
    "source": [
-    "icat.initialize(offline=True)"
+    "icat.initialize(offline=False)"
    ]
   },
   {
@@ -279,7 +294,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.12"
+   "version": "3.10.15"
   }
  },
  "nbformat": 4,