Fix notebook

examples/finetune_modernbert_on_glue.ipynb

@@ -42,7 +42,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "#! pip install transformers datasets accelerate scikit-learn -Uqq\n"
+    "#! pip install transformers datasets==2.15.0 accelerate scikit-learn -Uqq\n"
    ]
   },
   {
@@ -706,7 +706,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "When configuring `AutoModelForSequenceClassification`, two settings are critical to get things working with the HuggingFace `Trainer`. One is the `num_labels` we're expecting and the other is to set `compile=False` to avoid using the `torch.compile` function which is not supported in Transformers at the time of this writing."
+    "When configuring `AutoModelForSequenceClassification`, two settings are critical to get things working with the HuggingFace `Trainer`. One is the `num_labels` we're expecting and the other is to set `reference_compile=False` to avoid using the `torch.compile` function which is not supported in Transformers at the time of this writing."
    ]
   },
   {
@@ -725,7 +725,7 @@
    ],
    "source": [
     "hf_model = AutoModelForSequenceClassification.from_pretrained(\n",
-    "    checkpoint, num_labels=n_labels, id2label=id2label, label2id=label2id, compile=False\n",
+    "    checkpoint, num_labels=n_labels, id2label=id2label, label2id=label2id, reference_compile=False\n",
     ")\n"
    ]
   },
@@ -1167,7 +1167,7 @@
     "    # 5. Load the model and data collator\n",
     "    model_additional_kwargs = {\"id2label\": id2label, \"label2id\": label2id} if id2label and label2id else {}\n",
     "    hf_model = AutoModelForSequenceClassification.from_pretrained(\n",
-    "        checkpoint, num_labels=n_labels, compile=False, **model_additional_kwargs\n",
+    "        checkpoint, num_labels=n_labels, reference_compile=False, **model_additional_kwargs\n",
     "    )\n",
     "\n",
     "    hf_data_collator = DataCollatorWithPadding(tokenizer=hf_tokenizer)\n",

requirements-data.txt

@@ -1,4 +1,4 @@
-datasets
+datasets==2.15.0
 datatrove==0.2.0
 huggingface_hub==0.23.1
 pyyaml