[BugFix] Fix usage of BART in HF_Text

[sxjscience]

Issue #, if available:

Description of changes:

BART models won't support token_type_ids. This is identified via:

from transformers import AutoTokenizer

tokenizer_bart = AutoTokenizer.from_pretrained("facebook/bart-base")
print("BART: ", tokenizer_bart.model_input_names)

tokenizer_electra = AutoTokenizer.from_pretrained("google/electra-base-discriminator")
print("ELECTRA:", tokenizer_electra.model_input_names)

Output:

BART:  ['input_ids', 'attention_mask']
ELECTRA: ['input_ids', 'token_type_ids', 'attention_mask']

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Job PR-2787-63542a9 is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-2787/63542a9/index.html

[zhiqiangdon]

Should be "model.hf_text.checkpoint_name": checkpoint_name ?

[sxjscience]

@zhiqiangdon Just see this.. let me fix it in a follow-up PR about fixing the TIMM custom image size.

[zhiqiangdon]

LGTM. Left a comment about the test.

[cheungdaven]

LGTM