fix: special_tokens in the encode method to support the special token…

…s in the vocab

bpetokenizer/pretrained/wi17k_base/wi17k_base.json

@@ -17065,7 +17065,9 @@
         "(17306, 195)": 17307,
         "(17307, 163)": 17308,
         "(1012, 7365)": 17309,
-        "(9137, 336)": 17310
+        "(9137, 336)": 17310,
+        "(32, 32)": 17320,
+        "(17320, 32)": 17321
     },
     "vocab": {
         "0": "\\u0000",
@@ -34380,6 +34382,8 @@
         "17309": " differs",
         "17311": " def",
         "17312": "_stats",
-        "17313": " get"
+        "17313": " get",
+        "17320": "  ",
+        "17321": "   "
     }
 }

bpetokenizer/tokenizer.py

@@ -33,11 +33,13 @@ def __init__(self, pattern=None, special_tokens=None):
         self.special_tokens = {} if special_tokens is None else special_tokens
         self.inverse_special_tokens = {} if special_tokens is None else {v: k for k, v in special_tokens.items()}
         self.vocab_size = len(self.vocab) if self.vocab else 0
+        self.inverse_merges = {int(v): k for k, v in self.merges.items()} if self.merges else {}
 
     @classmethod
     def from_pretrained(cls, 
                         tokenizer_name: str, 
                         verbose=False):
+        """Allows you to load the pretrained tokenizers"""
         tokenizer = cls()
         pretrained_dir = 'bpetokenizer/pretrained'
         tokenizer_file = os.path.join(pretrained_dir, tokenizer_name, f'{tokenizer_name}.json')
@@ -60,6 +62,10 @@ def train(self, texts, vocab_size, verbose=False, min_frequency=1) -> None:
             vocab_size: int (the size of the vocab, gpt4 vocab size is around 100k)
             verbose: bool (to get extra visibilty and the overview of internal processes)
             min_frequency: int (the minimum frequency of the pair to be merged and added into the vocab as a new token)
+
+        internal_args:
+            text_chunks: list[str]
+            pair: tuple(int, int)
         """
         assert vocab_size >= 256
         num_merges = vocab_size - 256
@@ -150,33 +156,38 @@ def encode(self, text, special_tokens="none") -> list:
         else:
             raise ValueError(f"invalid special tokens argument: {special_tokens}")
 
-
-        text_chunks = re.findall(self.compiled_pattern, text)
+        if not special:
+            # shortcut: if no special tokens, just use the ordinary encoding
+            return self.encode_ord(text)
+        special_pattern = "(" + "|".join(re.escape(k) for k in special) + ")"
+        text_chunks = re.split(special_pattern, text)
         ids = []
         for chunk in text_chunks:
             if chunk in self.inverse_vocab:
                 ids.append(self.inverse_vocab[chunk])
-            elif chunk in self.special_tokens:
+            elif special and chunk in self.special_tokens:
                 ids.append(self.special_tokens[chunk])
             else:
                 chunk_ids = self._encode(chunk.encode("utf-8"))
                 ids.extend(chunk_ids)
         return ids
 
 
-    def decode(self, ids) -> str:
+    def decode(self, ids, verbose=False) -> str:
         part_bytes = []
         for idx in ids:
             if idx in self.vocab: #str conversion because vocab keys are strings when loaded from json
                 part_bytes.append(self.vocab[idx])
             elif idx in self.inverse_special_tokens:
                 part_bytes.append(self.inverse_special_tokens[idx].encode("utf-8")) # special tokens are not encoded in vocab
-            elif idx in self.merges:
-                pair = self.merges[idx]
+            elif idx in self.inverse_merges:
+                pair = self.inverse_merges[idx]
                 part_bytes.append(self.vocab[pair[0]] + self.vocab[pair[1]])
             else:
                 raise ValueError(f"invalid token id: {idx}")
         text_bytes = b"".join(part_bytes)
+        if verbose:
+            print("---\nText bytes: ", text_bytes)
         text = text_bytes.decode("utf-8", errors="replace")
         return text