mii/tokenizers
1
0
Fork 0
mirror of https://github.com/mii443/tokenizers.git synced 2025-08-22 16:25:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

Python - Improve documentation for post-processors

Browse Source
This commit is contained in:
Anthony MOI
2020-11-20 17:48:11 -05:00
committed by Anthony MOI
parent 933a2a9c99
commit 64441b54b1
2 changed files with 233 additions and 124 deletions
Download Patch File Download Diff File Expand all files Collapse all files

215
bindings/python/py_src/tokenizers/processors/__init__.pyi
View File

@ -10,13 +10,31 @@ class PostProcessor:
def num_special_tokens_to_add(self, is_pair):
"""
Return the number of special tokens that would be added for single/pair sentences.
:param is_pair: Boolean indicating if the input would be a single sentence or a pair
:return:
Args:
is_pair (:obj:`bool`):
Whether the input would be a pair of sequences
Returns:
:obj:`int`: The number of tokens to add
"""
pass
def process(self, encoding, pair=None, add_special_tokens=True):
"""
Post-process the given encodings, generating the final one
Args:
encoding (:class:`~tokenizers.Encoding`):
The encoding for the first sequence
pair (:class:`~tokenizers.Encoding`, `optional`):
The encoding for the pair sequence
add_special_tokens (:obj:`bool`):
Whether to add the special tokens
Return:
:class:`~tokenizers.Encoding`: The final encoding
"""
pass
@ -24,17 +42,16 @@ class BertProcessing(PostProcessor):
"""
This post-processor takes care of adding the special tokens needed by
a Bert model:
- a SEP token
- a CLS token
Args:
sep: Tuple[str, int]:
sep (:obj:`Tuple[str, int]`):
A tuple with the string representation of the SEP token, and its id
cls: Tuple[str, int]:
cls (:obj:`Tuple[str, int]`):
A tuple with the string representation of the CLS token, and its id
Returns:
PostProcessor
"""
def __init__(self, sep, cls):
@ -42,24 +59,43 @@ class BertProcessing(PostProcessor):
def num_special_tokens_to_add(self, is_pair):
"""
Return the number of special tokens that would be added for single/pair sentences.
:param is_pair: Boolean indicating if the input would be a single sentence or a pair
:return:
Args:
is_pair (:obj:`bool`):
Whether the input would be a pair of sequences
Returns:
:obj:`int`: The number of tokens to add
"""
pass
def process(self, encoding, pair=None, add_special_tokens=True):
"""
Post-process the given encodings, generating the final one
Args:
encoding (:class:`~tokenizers.Encoding`):
The encoding for the first sequence
pair (:class:`~tokenizers.Encoding`, `optional`):
The encoding for the pair sequence
add_special_tokens (:obj:`bool`):
Whether to add the special tokens
Return:
:class:`~tokenizers.Encoding`: The final encoding
"""
pass
class ByteLevel(PostProcessor):
"""
This post-processor takes care of trimming the offsets.
By default, the ByteLevel BPE might include whitespaces in the produced tokens. If you don't
want the offsets to include these whitespaces, then this PostProcessor must be used.
Args:
trim_offsets: bool:
trim_offsets (:obj:`bool`):
Whether to trim the whitespaces from the produced offsets.
"""
@ -68,13 +104,31 @@ class ByteLevel(PostProcessor):
def num_special_tokens_to_add(self, is_pair):
"""
Return the number of special tokens that would be added for single/pair sentences.
:param is_pair: Boolean indicating if the input would be a single sentence or a pair
:return:
Args:
is_pair (:obj:`bool`):
Whether the input would be a pair of sequences
Returns:
:obj:`int`: The number of tokens to add
"""
pass
def process(self, encoding, pair=None, add_special_tokens=True):
"""
Post-process the given encodings, generating the final one
Args:
encoding (:class:`~tokenizers.Encoding`):
The encoding for the first sequence
pair (:class:`~tokenizers.Encoding`, `optional`):
The encoding for the pair sequence
add_special_tokens (:obj:`bool`):
Whether to add the special tokens
Return:
:class:`~tokenizers.Encoding`: The final encoding
"""
pass
@ -82,29 +136,28 @@ class RobertaProcessing(PostProcessor):
"""
This post-processor takes care of adding the special tokens needed by
a Roberta model:
- a SEP token
- a CLS token
It also takes care of trimming the offsets.
By default, the ByteLevel BPE might include whitespaces in the produced tokens. If you don't
want the offsets to include these whitespaces, then this PostProcessor should be initialized
with `trim_offsets=True`
with :obj:`trim_offsets=True`
Args:
sep: Tuple[str, int]:
sep (:obj:`Tuple[str, int]`):
A tuple with the string representation of the SEP token, and its id
cls: Tuple[str, int]:
cls (:obj:`Tuple[str, int]`):
A tuple with the string representation of the CLS token, and its id
trim_offsets: bool:
trim_offsets (:obj:`bool`, `optional`, defaults to :obj:`True`):
Whether to trim the whitespaces from the produced offsets.
add_prefix_space: bool:
add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
Whether the add_prefix_space option was enabled during pre-tokenization. This
is relevant because it defines the way the offsets are trimmed out.
Returns:
PostProcessor
"""
def __init__(self, sep, cls, trim_offsets=True, add_prefix_space=True):
@ -112,13 +165,31 @@ class RobertaProcessing(PostProcessor):
def num_special_tokens_to_add(self, is_pair):
"""
Return the number of special tokens that would be added for single/pair sentences.
:param is_pair: Boolean indicating if the input would be a single sentence or a pair
:return:
Args:
is_pair (:obj:`bool`):
Whether the input would be a pair of sequences
Returns:
:obj:`int`: The number of tokens to add
"""
pass
def process(self, encoding, pair=None, add_special_tokens=True):
"""
Post-process the given encodings, generating the final one
Args:
encoding (:class:`~tokenizers.Encoding`):
The encoding for the first sequence
pair (:class:`~tokenizers.Encoding`, `optional`):
The encoding for the pair sequence
add_special_tokens (:obj:`bool`):
Whether to add the special tokens
Return:
:class:`~tokenizers.Encoding`: The final encoding
"""
pass
@ -127,35 +198,36 @@ class TemplateProcessing(PostProcessor):
Provides a way to specify templates in order to add the special tokens to each
input sequence as relevant.
Let's take `BERT` tokenizer as an example. It uses two special tokens, used to
delimitate each sequence. `[CLS]` is always used at the beginning of the first
sequence, and `[SEP]` is added at the end of both the first, and the pair
Let's take :obj:`BERT` tokenizer as an example. It uses two special tokens, used to
delimitate each sequence. :obj:`[CLS]` is always used at the beginning of the first
sequence, and :obj:`[SEP]` is added at the end of both the first, and the pair
sequences. The final result looks like this:
- Single sequence: `[CLS] Hello there [SEP]`
- Pair sequences: `[CLS] My name is Anthony [SEP] What is my name? [SEP]`
With the type ids as following:
```markdown
[CLS] ... [SEP] ... [SEP]
0 0 0 1 1
```
You can achieve such behavior using a TemplateProcessing:
```
TemplateProcessing(
single="[CLS] $0 [SEP]",
pair="[CLS] $A [SEP] $B:1 [SEP]:1",
special_tokens=[("[CLS]", 1), ("[SEP]", 0)],
)
```
- Single sequence: :obj:`[CLS] Hello there [SEP]`
- Pair sequences: :obj:`[CLS] My name is Anthony [SEP] What is my name? [SEP]`
In this example, each input sequence is identified using a `$` construct. This identifier
With the type ids as following::
[CLS] ... [SEP] ... [SEP]
0 0 0 1 1
You can achieve such behavior using a TemplateProcessing::
TemplateProcessing(
single="[CLS] $0 [SEP]",
pair="[CLS] $A [SEP] $B:1 [SEP]:1",
special_tokens=[("[CLS]", 1), ("[SEP]", 0)],
)
In this example, each input sequence is identified using a ``$`` construct. This identifier
lets us specify each input sequence, and the type_id to use. When nothing is specified,
it uses the default values. Here are the different ways to specify it:
- Specifying the sequence, with default `type_id == 0`: `$A` or `$B`
- Specifying the `type_id` with default `sequence == A`: `$0`, `$1`, `$2`, ...
- Specifying both: `$A:0`, `$B:1`, ...
The same construct is used for special tokens: `<identifier>(:<type_id>)?`.
- Specifying the sequence, with default ``type_id == 0``: ``$A`` or ``$B``
- Specifying the `type_id` with default ``sequence == A``: ``$0``, ``$1``, ``$2``, ...
- Specifying both: ``$A:0``, ``$B:1``, ...
The same construct is used for special tokens: ``<identifier>(:<type_id>)?``.
**Warning**: You must ensure that you are giving the correct tokens/ids as these
will be added to the Encoding without any further check. If the given ids correspond
@ -163,27 +235,30 @@ class TemplateProcessing(PostProcessor):
might lead to unexpected results.
Args:
single: Template
single (:obj:`Template`):
The template used for single sequences
pair: Template:
pair (:obj:`Template`):
The template used when both sequences are specified
special_tokens: Tokens:
special_tokens (:obj:`Tokens`):
The list of special tokens used in each sequences
Template: Union[str, List[str]]:
- If a `str` is provided, the whitespace is used as delimiter between tokens
- If a `List[str]` is provided, a list of tokens
Types:
Tokens: List[Union[Tuple[int, str], Tuple[str, int], dict]]:
- A Tuple with both a token and its associated ID, in any order
- A dict with the following keys:
- "id": str => The special token id, as specified in the Template
- "ids": List[int] => The associated IDs
- "tokens": List[str] => The associated tokens
The given dict expects the provided `ids` and `tokens` lists to have
the same length.
Template (:obj:`str` or :obj:`List`):
- If a :obj:`str` is provided, the whitespace is used as delimiter between tokens
- If a :obj:`List[str]` is provided, a list of tokens
Tokens (:obj:`List[Union[Tuple[int, str], Tuple[str, int], dict]]`):
- A :obj:`Tuple` with both a token and its associated ID, in any order
- A :obj:`dict` with the following keys:
- "id": :obj:`str` => The special token id, as specified in the Template
- "ids": :obj:`List[int]` => The associated IDs
- "tokens": :obj:`List[str]` => The associated tokens
The given dict expects the provided :obj:`ids` and :obj:`tokens` lists to have
the same length.
"""
def __init__(self, single, pair, special_tokens):
@ -191,12 +266,30 @@ class TemplateProcessing(PostProcessor):
def num_special_tokens_to_add(self, is_pair):
"""
Return the number of special tokens that would be added for single/pair sentences.
:param is_pair: Boolean indicating if the input would be a single sentence or a pair
:return:
Args:
is_pair (:obj:`bool`):
Whether the input would be a pair of sequences
Returns:
:obj:`int`: The number of tokens to add
"""
pass
def process(self, encoding, pair=None, add_special_tokens=True):
"""
Post-process the given encodings, generating the final one
Args:
encoding (:class:`~tokenizers.Encoding`):
The encoding for the first sequence
pair (:class:`~tokenizers.Encoding`, `optional`):
The encoding for the pair sequence
add_special_tokens (:obj:`bool`):
Whether to add the special tokens
Return:
:class:`~tokenizers.Encoding`: The final encoding
"""
pass

142
bindings/python/src/processors.rs
View File

@ -93,14 +93,32 @@ impl PyPostProcessor {
}
/// Return the number of special tokens that would be added for single/pair sentences.
/// :param is_pair: Boolean indicating if the input would be a single sentence or a pair
/// :return:
///
/// Args:
/// is_pair (:obj:`bool`):
/// Whether the input would be a pair of sequences
///
/// Returns:
/// :obj:`int`: The number of tokens to add
#[text_signature = "(self, is_pair)"]
fn num_special_tokens_to_add(&self, is_pair: bool) -> usize {
self.processor.added_tokens(is_pair)
}
/// Post-process the given encodings, generating the final one
///
/// Args:
/// encoding (:class:`~tokenizers.Encoding`):
/// The encoding for the first sequence
///
/// pair (:class:`~tokenizers.Encoding`, `optional`):
/// The encoding for the pair sequence
///
/// add_special_tokens (:obj:`bool`):
/// Whether to add the special tokens
///
/// Return:
/// :class:`~tokenizers.Encoding`: The final encoding
#[args(pair = "None", add_special_tokens = "true")]
#[text_signature = "(self, encoding, pair=None, add_special_tokens=True)"]
fn process(
@ -121,17 +139,16 @@ impl PyPostProcessor {
/// This post-processor takes care of adding the special tokens needed by
/// a Bert model:
///
/// - a SEP token
/// - a CLS token
///
/// Args:
/// sep: Tuple[str, int]:
/// sep (:obj:`Tuple[str, int]`):
/// A tuple with the string representation of the SEP token, and its id
///
/// cls: Tuple[str, int]:
/// cls (:obj:`Tuple[str, int]`):
/// A tuple with the string representation of the CLS token, and its id
///
/// Returns:
/// PostProcessor
#[pyclass(extends=PyPostProcessor, module = "tokenizers.processors", name=BertProcessing)]
#[text_signature = "(self, sep, cls)"]
pub struct PyBertProcessing {}
@ -152,29 +169,28 @@ impl PyBertProcessing {
/// This post-processor takes care of adding the special tokens needed by
/// a Roberta model:
///
/// - a SEP token
/// - a CLS token
///
/// It also takes care of trimming the offsets.
/// By default, the ByteLevel BPE might include whitespaces in the produced tokens. If you don't
/// want the offsets to include these whitespaces, then this PostProcessor should be initialized
/// with `trim_offsets=True`
/// with :obj:`trim_offsets=True`
///
/// Args:
/// sep: Tuple[str, int]:
/// sep (:obj:`Tuple[str, int]`):
/// A tuple with the string representation of the SEP token, and its id
///
/// cls: Tuple[str, int]:
/// cls (:obj:`Tuple[str, int]`):
/// A tuple with the string representation of the CLS token, and its id
///
/// trim_offsets: bool:
/// trim_offsets (:obj:`bool`, `optional`, defaults to :obj:`True`):
/// Whether to trim the whitespaces from the produced offsets.
///
/// add_prefix_space: bool:
/// add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
/// Whether the add_prefix_space option was enabled during pre-tokenization. This
/// is relevant because it defines the way the offsets are trimmed out.
///
/// Returns:
/// PostProcessor
#[pyclass(extends=PyPostProcessor, module = "tokenizers.processors", name=RobertaProcessing)]
#[text_signature = "(self, sep, cls, trim_offsets=True, add_prefix_space=True)"]
pub struct PyRobertaProcessing {}
@ -203,11 +219,12 @@ impl PyRobertaProcessing {
}
/// This post-processor takes care of trimming the offsets.
///
/// By default, the ByteLevel BPE might include whitespaces in the produced tokens. If you don't
/// want the offsets to include these whitespaces, then this PostProcessor must be used.
///
/// Args:
/// trim_offsets: bool:
/// trim_offsets (:obj:`bool`):
/// Whether to trim the whitespaces from the produced offsets.
#[pyclass(extends=PyPostProcessor, module = "tokenizers.processors", name=ByteLevel)]
#[text_signature = "(self, trim_offsets=True)"]
@ -215,19 +232,14 @@ pub struct PyByteLevel {}
#[pymethods]
impl PyByteLevel {
#[new]
#[args(kwargs = "**")]
fn new(kwargs: Option<&PyDict>) -> PyResult<(Self, PyPostProcessor)> {
#[args(trim_offsets = "None")]
fn new(trim_offsets: Option<bool>) -> PyResult<(Self, PyPostProcessor)> {
let mut byte_level = ByteLevel::default();
if let Some(kwargs) = kwargs {
for (key, value) in kwargs {
let key: &str = key.extract()?;
match key {
"trim_offsets" => byte_level = byte_level.trim_offsets(value.extract()?),
_ => println!("Ignored unknown kwargs option {}", key),
}
}
if let Some(to) = trim_offsets {
byte_level = byte_level.trim_offsets(to);
}
Ok((
PyByteLevel {},
PyPostProcessor::new(Arc::new(byte_level.into())),
@ -305,35 +317,36 @@ impl FromPyObject<'_> for PyTemplate {
/// Provides a way to specify templates in order to add the special tokens to each
/// input sequence as relevant.
///
/// Let's take `BERT` tokenizer as an example. It uses two special tokens, used to
/// delimitate each sequence. `[CLS]` is always used at the beginning of the first
/// sequence, and `[SEP]` is added at the end of both the first, and the pair
/// Let's take :obj:`BERT` tokenizer as an example. It uses two special tokens, used to
/// delimitate each sequence. :obj:`[CLS]` is always used at the beginning of the first
/// sequence, and :obj:`[SEP]` is added at the end of both the first, and the pair
/// sequences. The final result looks like this:
/// - Single sequence: `[CLS] Hello there [SEP]`
/// - Pair sequences: `[CLS] My name is Anthony [SEP] What is my name? [SEP]`
/// With the type ids as following:
/// ```markdown
/// [CLS] ... [SEP] ... [SEP]
/// 0 0 0 1 1
/// ```
///
/// You can achieve such behavior using a TemplateProcessing:
/// ```
/// TemplateProcessing(
/// single="[CLS] $0 [SEP]",
/// pair="[CLS] $A [SEP] $B:1 [SEP]:1",
/// special_tokens=[("[CLS]", 1), ("[SEP]", 0)],
/// )
/// ```
/// - Single sequence: :obj:`[CLS] Hello there [SEP]`
/// - Pair sequences: :obj:`[CLS] My name is Anthony [SEP] What is my name? [SEP]`
///
/// In this example, each input sequence is identified using a `$` construct. This identifier
/// With the type ids as following::
///
/// [CLS] ... [SEP] ... [SEP]
/// 0 0 0 1 1
///
/// You can achieve such behavior using a TemplateProcessing::
///
/// TemplateProcessing(
/// single="[CLS] $0 [SEP]",
/// pair="[CLS] $A [SEP] $B:1 [SEP]:1",
/// special_tokens=[("[CLS]", 1), ("[SEP]", 0)],
/// )
///
/// In this example, each input sequence is identified using a ``$`` construct. This identifier
/// lets us specify each input sequence, and the type_id to use. When nothing is specified,
/// it uses the default values. Here are the different ways to specify it:
/// - Specifying the sequence, with default `type_id == 0`: `$A` or `$B`
/// - Specifying the `type_id` with default `sequence == A`: `$0`, `$1`, `$2`, ...
/// - Specifying both: `$A:0`, `$B:1`, ...
///
/// The same construct is used for special tokens: `<identifier>(:<type_id>)?`.
/// - Specifying the sequence, with default ``type_id == 0``: ``$A`` or ``$B``
/// - Specifying the `type_id` with default ``sequence == A``: ``$0``, ``$1``, ``$2``, ...
/// - Specifying both: ``$A:0``, ``$B:1``, ...
///
/// The same construct is used for special tokens: ``<identifier>(:<type_id>)?``.
///
/// **Warning**: You must ensure that you are giving the correct tokens/ids as these
/// will be added to the Encoding without any further check. If the given ids correspond
@ -341,27 +354,30 @@ impl FromPyObject<'_> for PyTemplate {
/// might lead to unexpected results.
///
/// Args:
/// single: Template
/// single (:obj:`Template`):
/// The template used for single sequences
///
/// pair: Template:
/// pair (:obj:`Template`):
/// The template used when both sequences are specified
///
/// special_tokens: Tokens:
/// special_tokens (:obj:`Tokens`):
/// The list of special tokens used in each sequences
///
/// Template: Union[str, List[str]]:
/// - If a `str` is provided, the whitespace is used as delimiter between tokens
/// - If a `List[str]` is provided, a list of tokens
/// Types:
///
/// Tokens: List[Union[Tuple[int, str], Tuple[str, int], dict]]:
/// - A Tuple with both a token and its associated ID, in any order
/// - A dict with the following keys:
/// - "id": str => The special token id, as specified in the Template
/// - "ids": List[int] => The associated IDs
/// - "tokens": List[str] => The associated tokens
/// The given dict expects the provided `ids` and `tokens` lists to have
/// the same length.
/// Template (:obj:`str` or :obj:`List`):
/// - If a :obj:`str` is provided, the whitespace is used as delimiter between tokens
/// - If a :obj:`List[str]` is provided, a list of tokens
///
/// Tokens (:obj:`List[Union[Tuple[int, str], Tuple[str, int], dict]]`):
/// - A :obj:`Tuple` with both a token and its associated ID, in any order
/// - A :obj:`dict` with the following keys:
/// - "id": :obj:`str` => The special token id, as specified in the Template
/// - "ids": :obj:`List[int]` => The associated IDs
/// - "tokens": :obj:`List[str]` => The associated tokens
///
/// The given dict expects the provided :obj:`ids` and :obj:`tokens` lists to have
/// the same length.
#[pyclass(extends=PyPostProcessor, module = "tokenizers.processors", name=TemplateProcessing)]
#[text_signature = "(self, single, pair, special_tokens)"]
pub struct PyTemplateProcessing {}