remove resize token embeddings code

djliden · Mar 8, 2024 · 42d53d2 · 42d53d2
1 parent b0ff299
commit 42d53d2
Showing 1 changed file with 5 additions and 2 deletions.
diff --git a/notebooks/4_olmo_1b_instruction_tune/4_olmo_instruction_tune.ipynb b/notebooks/4_olmo_1b_instruction_tune/4_olmo_instruction_tune.ipynb
@@ -577,8 +577,11 @@
     "special_tokens = [\"<|im_start|>\", \"<|im_end|>\"]\n",
     "# Adding special tokens to the tokenizer\n",
     "tokenizer.add_special_tokens({\"additional_special_tokens\": special_tokens})\n",
-    "# Resize the model's input token embeddings matrix\n",
-    "model.resize_token_embeddings(len(tokenizer))\n",
+    "\n",
+    "# Do not need to resize the model's input token embeddings matrix\n",
+    "# it is already larger than the vocabulary/large enough to accommodate\n",
+    "# the added tokens\n",
+    "# model.resize_token_embeddings(len(tokenizer))\n",
     "\n",
     "def format_slimorca(ex):\n",
     "    role_mapping = {\"gpt\": \"assistant\", \"system\": \"system\", \"human\": \"user\"}\n",