Fixing doc. (#1499)

* Fixing doc.

* SentencePieceUnigram  and Convert.py still used sentencepiece

* stub

---------

Co-authored-by: Arthur Zucker <arthur.zucker@gmail.com>

bindings/python/py_src/tokenizers/decoders/__init__.pyi

@@ -152,9 +152,11 @@ class Metaspace(Decoder):
             The replacement character. Must be exactly one character. By default we
             use the `▁` (U+2581) meta symbol (Same as in SentencePiece).
 
-        add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
+        prepend_scheme (:obj:`str`, `optional`, defaults to :obj:`"always"`):
             Whether to add a space to the first word if there isn't already one. This
             lets us treat `hello` exactly like `say hello`.
+            Choices: "always", "never", "first". First means the space is only added on the first
+            token (relevant when special tokens are used or other pre_tokenizer are used).
     """
     def __init__(self, replacement="▁", prepend_scheme="always", split=True):
         pass

bindings/python/py_src/tokenizers/implementations/sentencepiece_unigram.py

@@ -183,8 +183,9 @@ def from_spm(filename: str):
             )
         else:
             tokenizer.normalizer = normalizers.Sequence([normalizers.Replace(Regex(" {2,}"), " ")])
-        tokenizer.pre_tokenizer = pre_tokenizers.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
-        tokenizer.decoder = decoders.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
+        prepend_scheme = "always" if add_prefix_space else "never"
+        tokenizer.pre_tokenizer = pre_tokenizers.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme)
+        tokenizer.decoder = decoders.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme)
 
         parameters = {
             "model": "SentencePieceUnigram",

bindings/python/py_src/tokenizers/pre_tokenizers/__init__.pyi

@@ -270,9 +270,12 @@ class Metaspace(PreTokenizer):
             The replacement character. Must be exactly one character. By default we
             use the `▁` (U+2581) meta symbol (Same as in SentencePiece).
 
-        add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
+        prepend_scheme (:obj:`str`, `optional`, defaults to :obj:`"always"`):
             Whether to add a space to the first word if there isn't already one. This
             lets us treat `hello` exactly like `say hello`.
+            Choices: "always", "never", "first". First means the space is only added on the first
+            token (relevant when special tokens are used or other pre_tokenizer are used).
+
     """
     def __init__(self, replacement="_", prepend_scheme="always", split=True):
         pass

bindings/python/scripts/convert.py

@@ -102,9 +102,9 @@ def converted(self):
         tokenizer.normalizer = self.normalizer(self.proto)
 
         replacement = "▁"
-        add_prefix_space = True
-        tokenizer.pre_tokenizer = Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
-        tokenizer.decoder = decoders.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
+        prepend_scheme = "always"
+        tokenizer.pre_tokenizer = Metaspace(replacement=replacement, prepend_scheme=prepend_scheme)
+        tokenizer.decoder = decoders.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme)
         post_processor = self.post_processor(tokenizer)
         if post_processor:
             tokenizer.post_processor = post_processor

bindings/python/src/decoders.rs

@@ -304,9 +304,11 @@ impl PyStrip {
 ///         The replacement character. Must be exactly one character. By default we
 ///         use the `▁` (U+2581) meta symbol (Same as in SentencePiece).
 ///
-///     add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
+///     prepend_scheme (:obj:`str`, `optional`, defaults to :obj:`"always"`):
 ///         Whether to add a space to the first word if there isn't already one. This
 ///         lets us treat `hello` exactly like `say hello`.
+///         Choices: "always", "never", "first". First means the space is only added on the first
+///         token (relevant when special tokens are used or other pre_tokenizer are used).
 #[pyclass(extends=PyDecoder, module = "tokenizers.decoders", name = "Metaspace")]
 pub struct PyMetaspaceDec {}
 #[pymethods]

bindings/python/src/pre_tokenizers.rs

@@ -477,9 +477,12 @@ pub(crate) fn from_string(string: String) -> Result<PrependScheme, PyErr> {
 ///         The replacement character. Must be exactly one character. By default we
 ///         use the `▁` (U+2581) meta symbol (Same as in SentencePiece).
 ///
-///     add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
+///     prepend_scheme (:obj:`str`, `optional`, defaults to :obj:`"always"`):
 ///         Whether to add a space to the first word if there isn't already one. This
 ///         lets us treat `hello` exactly like `say hello`.
+///         Choices: "always", "never", "first". First means the space is only added on the first
+///         token (relevant when special tokens are used or other pre_tokenizer are used).
+///
 #[pyclass(extends=PyPreTokenizer, module = "tokenizers.pre_tokenizers", name = "Metaspace")]
 pub struct PyMetaspace {}
 #[pymethods]

bindings/python/tests/bindings/test_pre_tokenizers.py

@@ -109,8 +109,6 @@ def test_can_modify(self):
         # Modify these
         pretok.replacement = "%"
         assert pretok.replacement == "%"
-        pretok.add_prefix_space = True
-        assert pretok.add_prefix_space == True
         pretok.prepend_scheme = "first"
         assert pretok.prepend_scheme == "first"
         pretok.split = True