[GPT2] Add correct keys on _keys_to_ignore_on_load_unexpected on all child classes of GPT2PreTrainedModel

[younesbelkada]

What does this PR do?

as per title

forgot to add them in #23256

Currently this snippet:

from transformers import GPT2Model

model = GPT2Model.from_pretrained("gpt2")

Gives a big warning:

Some weights of the model checkpoint at gpt2 were not used when initializing GPT2Model: ['h.10.attn.bias', 'h.5.attn.bias', 'h.7.attn.bias', 'h.0.attn.bias', 'h.11.attn.bias', 'h.8.attn.bias', 'h.1.attn.bias', 'h.9.attn.bias', 'h.2.attn.bias', 'h.4.attn.bias', 'h.6.attn.bias', 'h.3.attn.bias']
- This IS expected if you are initializing GPT2Model from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing GPT2Model from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).

This PR fixes it by adding the correct regex expressions on _keys_to_ignore_on_load_unexpected for all child classes that inherit from GPT2PreTrainedModel

cc @sgugger @patrickvonplaten

[sgugger]

Thanks!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.