Fixing OPT fast tokenizer option. (#18753)

* Fixing OPT fast tokenizer option.

* Remove dependency on `pt`.

* Move it to GPT2 tokenization tests.

* Added a few tests.

src/transformers/convert_slow_tokenizer.py

@@ -282,8 +282,20 @@ def converted(self) -> Tokenizer:
 
         tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=self.original_tokenizer.add_prefix_space)
         tokenizer.decoder = decoders.ByteLevel()
-        tokenizer.post_processor = processors.ByteLevel(trim_offsets=False)
-
+        if self.original_tokenizer.add_bos_token:
+            bos = self.original_tokenizer.bos_token
+            bos_token_id = self.original_tokenizer.bos_token_id
+            tokenizer.post_processor = processors.TemplateProcessing(
+                single=f"{bos}:0 $A:0",  # token_type_id is 2 for Funnel transformer
+                pair=f"{bos}:0 $A:0 $B:1",
+                special_tokens=[
+                    (bos, bos_token_id),
+                ],
+            )
+        else:
+            # XXX trim_offsets=False actually means this post_processor doesn't
+            # really do anything.
+            tokenizer.post_processor = processors.ByteLevel(trim_offsets=False)
         return tokenizer
 
 

src/transformers/models/gpt2/tokenization_gpt2_fast.py

@@ -146,16 +146,7 @@ def __init__(
             **kwargs,
         )
 
-        if kwargs.pop("add_bos_token", False):
-            model_id = kwargs.pop("name_or_path", "")
-            raise ValueError(
-                "Currenty GPT2's fast tokenizer does NOT support adding a BOS token."
-                "Instead you should use GPT2's slow tokenizer class `GPT2Tokenizer` as follows: \n"
-                f"`GPT2Tokenizer.from_pretrained('{model_id}')`\nor\n"
-                f"`AutoTokenizer.from_pretrained('{model_id}', use_fast=False)`\n"
-                "This issue will be fixed soon, see: https://github.com/huggingface/tokenizers/pull/1005."
-                " so that the fast tokenizer works correctly."
-            )
+        self.add_bos_token = kwargs.pop("add_bos_token", False)
 
         pre_tok_state = json.loads(self.backend_tokenizer.pre_tokenizer.__getstate__())
         if pre_tok_state.get("add_prefix_space", add_prefix_space) != add_prefix_space:

tests/models/gpt2/test_tokenization_gpt2.py

@@ -18,7 +18,7 @@
 import os
 import unittest
 
-from transformers import GPT2Tokenizer, GPT2TokenizerFast
+from transformers import AutoTokenizer, GPT2Tokenizer, GPT2TokenizerFast
 from transformers.models.gpt2.tokenization_gpt2 import VOCAB_FILES_NAMES
 from transformers.testing_utils import require_tokenizers
 
@@ -275,3 +275,57 @@ def test_special_tokens_mask_input_pairs_and_bos_token(self):
                 ]
                 filtered_sequence = [x for x in filtered_sequence if x is not None]
                 self.assertEqual(encoded_sequence, filtered_sequence)
+
+
+@require_tokenizers
+class OPTTokenizationTest(unittest.TestCase):
+    def test_serialize_deserialize_fast_opt(self):
+        # More context:
+        # https://huggingface.co/wjmcat/opt-350m-paddle/discussions/1
+        # https://huggingface.slack.com/archives/C01N44FJDHT/p1653511495183519
+        # https://github.com/huggingface/transformers/pull/17088#discussion_r871246439
+
+        tokenizer = AutoTokenizer.from_pretrained("facebook/opt-350m", from_slow=True)
+        text = "A photo of a cat"
+
+        tokens_ids = tokenizer.encode(
+            text,
+        )
+        self.assertEqual(tokens_ids, [2, 250, 1345, 9, 10, 4758])
+        tokenizer.save_pretrained("test_opt")
+
+        tokenizer = AutoTokenizer.from_pretrained("./test_opt")
+        tokens_ids = tokenizer.encode(
+            text,
+        )
+        self.assertEqual(tokens_ids, [2, 250, 1345, 9, 10, 4758])
+
+    def test_fast_slow_equivalence(self):
+        tokenizer = AutoTokenizer.from_pretrained("facebook/opt-350m", use_slow=True)
+        text = "A photo of a cat"
+
+        tokens_ids = tokenizer.encode(
+            text,
+        )
+        # Same as above
+        self.assertEqual(tokens_ids, [2, 250, 1345, 9, 10, 4758])
+
+    def test_users_can_modify_bos(self):
+        tokenizer = AutoTokenizer.from_pretrained("facebook/opt-350m", from_slow=True)
+
+        tokenizer.bos_token = "bos"
+        tokenizer.bos_token_id = tokenizer.get_vocab()["bos"]
+
+        text = "A photo of a cat"
+        tokens_ids = tokenizer.encode(
+            text,
+        )
+        # We changed the bos token
+        self.assertEqual(tokens_ids, [31957, 250, 1345, 9, 10, 4758])
+        tokenizer.save_pretrained("./tok")
+        tokenizer = AutoTokenizer.from_pretrained("./tok")
+        self.assertTrue(tokenizer.is_fast)
+        tokens_ids = tokenizer.encode(
+            text,
+        )
+        self.assertEqual(tokens_ids, [31957, 250, 1345, 9, 10, 4758])