Fix tokenization in SQuAD for RoBERTa, Longformer, BART (#7387)

* fix squad tokenization for roberta & co

* change to pure type based check

* sort imports

src/transformers/data/processors/squad.py

@@ -7,7 +7,10 @@ import numpy as np
 from tqdm import tqdm
 
 from ...file_utils import is_tf_available, is_torch_available
+from ...tokenization_bart import BartTokenizer
 from ...tokenization_bert import whitespace_tokenize
+from ...tokenization_longformer import LongformerTokenizer
+from ...tokenization_roberta import RobertaTokenizer
 from ...tokenization_utils_base import TruncationStrategy
 from ...utils import logging
 from .utils import DataProcessor
@@ -109,6 +112,9 @@ def squad_convert_example_to_features(
     all_doc_tokens = []
     for (i, token) in enumerate(example.doc_tokens):
         orig_to_tok_index.append(len(all_doc_tokens))
+        if isinstance(tokenizer, (RobertaTokenizer, LongformerTokenizer, BartTokenizer)):
+            sub_tokens = tokenizer.tokenize(token, add_prefix_space=True)
+        else:
             sub_tokens = tokenizer.tokenize(token)
         for sub_token in sub_tokens:
             tok_to_orig_index.append(i)