Adding support for tokens being suffixes or part of each other. (#13918)

* Adding support for tokens being suffixes or part of each other.

* Better test name.

src/transformers/tokenization_utils.py

@@ -150,26 +150,44 @@ class Trie:
 
                     # Lookahead to match longest first
                     # Important in case of extra_id_1 vs extra_id_100
-                    lookahead_index = current
-                    end = current
-                    next_char = text[lookahead_index] if lookahead_index < len(text) else None
-                    while next_char in trie_pointer:
-                        trie_pointer = trie_pointer[next_char]
-                        lookahead_index += 1
-                        if "" in trie_pointer:
-                            end = lookahead_index
-                            skip = lookahead_index
-
-                        if lookahead_index == len(text):
-                            # End of string
+                    # Here we are also actively looking for other earlier partial
+                    # matches
+                    # "[CLS]", "L", we need to match CLS even if L is special
+                    for lookstart, looktrie_pointer in states.items():
+                        if lookstart > start:
+                            # This partial match is later, we can stop looking
                             break
-                        next_char = text[lookahead_index]
-                    # End lookahead
+                        elif lookstart < start:
+                            # This partial match is earlier, the trie pointer
+                            # was already updated, so index is + 1
+                            lookahead_index = current + 1
+                            end = current + 1
+                        else:
+                            # Here lookstart == start and
+                            #      looktrie_pointer == trie_pointer
+                            # It wasn't updated yet so indices are current ones
+                            lookahead_index = current
+                            end = current
+                        next_char = text[lookahead_index] if lookahead_index < len(text) else None
+                        while next_char in looktrie_pointer:
+                            looktrie_pointer = looktrie_pointer[next_char]
+                            lookahead_index += 1
+                            if "" in looktrie_pointer:
+                                start = lookstart
+                                end = lookahead_index
+                                skip = lookahead_index
+
+                            if lookahead_index == len(text):
+                                # End of string
+                                break
+                            next_char = text[lookahead_index]
+                        # End lookahead
 
                     # Storing and resetting
                     offsets.append(start)
                     offsets.append(end)
                     reset = True
+                    break
                 elif current_char in trie_pointer:
                     # The current character being looked at has a match within the trie
                     # update the pointer (it will be stored back into states later).
@@ -210,6 +228,9 @@ class Trie:
                 # item so we need to break.
                 break
 
+        return self.cut_text(text, offsets)
+
+    def cut_text(self, text, offsets):
         # We have all the offsets now, we just need to do the actual splitting.
         # We need to eventually add the first part of the string and the eventual
         # last part.
@@ -217,7 +238,12 @@ class Trie:
         tokens = []
         start = 0
         for end in offsets:
-            if start == end:
+            if start > end:
+                logger.error(
+                    "There was a bug in Trie algorithm in tokenization. Attempting to recover. Please report it anyway."
+                )
+                continue
+            elif start == end:
                 # This might happen if there's a match at index 0
                 # we're also preventing zero-width cuts in case of two
                 # consecutive matches

tests/test_tokenization_common.py

@@ -3574,3 +3574,24 @@ class TrieTest(unittest.TestCase):
         trie.add("TOKEN]")
         trie.add("[SPECIAL_TOKEN]")
         self.assertEqual(trie.split("This is something [SPECIAL_TOKEN]"), ["This is something ", "[SPECIAL_TOKEN]"])
+
+    def test_trie_subtokens(self):
+        trie = Trie()
+        trie.add("A")
+        trie.add("P")
+        trie.add("[SPECIAL_TOKEN]")
+        self.assertEqual(trie.split("This is something [SPECIAL_TOKEN]"), ["This is something ", "[SPECIAL_TOKEN]"])
+
+    def test_trie_suffix_tokens(self):
+        trie = Trie()
+        trie.add("AB")
+        trie.add("B")
+        trie.add("C")
+        self.assertEqual(trie.split("ABC"), ["AB", "C"])
+
+    def test_cut_text_hardening(self):
+        # Even if the offsets are wrong, we necessarily output correct string
+        # parts.
+        trie = Trie()
+        parts = trie.cut_text("ABC", [0, 0, 2, 1, 2, 3])
+        self.assertEqual(parts, ["AB", "C"])