Binary pretrained model can't train a multi-class classifier?

I use model_checkpoint = "clefourrier/graphormer-base-pcqm4mv2" # pre-trained model from which to fine-tune to be my pretrained model, and my dataset is here.
I follow the code in this website.
But change the

model = GraphormerForGraphClassification.from_pretrained(
    model_checkpoint, 

    # We have 167 attack patterns and 1 benign
    num_classes=168, 

    # provide this in case you're planning to fine-tune an already fine-tuned checkpoint
    ignore_mismatched_sizes = True, 

Since I have 168 class to be classified.
Then when I run the python script, I encountered:

Some weights of GraphormerForGraphClassification were not initialized from the model checkpoint at clefourrier/graphormer-base-pcqm4mv2 and are newly initialized because the shapes did not match:
- classifier.classifier.weight: found shape torch.Size([1, 768]) in the checkpoint and torch.Size([168, 768]) in the model instantiated
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
  0%|                                                                                                                                                                                                                                          | 0/23120 [00:00<?, ?it/s]

For about 20 min, the log didn't move at all.

Start with the GraphormerForGraphClassification.from_pretrained call and the graphml-classification.md example, then reproduce the Python script using the clefourrier/graphormer-base-pcqm4mv2 checkpoint and 168 classes. Check whether the classifier-size warning or the stalled training progress is the issue to address; done means the multi-class training run proceeds or the failure is clearly explained.