mii/tokenizers
1
0
Fork 0
mirror of https://github.com/mii443/tokenizers.git synced 2025-08-22 08:15:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix pylist (#1673)

Browse Source 
* fix pylist

* add comment about why we use PySequence

* style

* fix encode batch fast as well

* Update bindings/python/src/tokenizer.rs

Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>

* fix with capacity

* stub :)

---------

Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
This commit is contained in:
Arthur
2024-11-05 16:24:23 +01:00
committed by GitHub
parent 0f3a3f957e
commit 5e223ceb48
2 changed files with 30 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
bindings/python/py_src/tokenizers/__init__.pyi
View File

@ -859,7 +859,9 @@ class Tokenizer:
def encode_batch(self, input, is_pretokenized=False, add_special_tokens=True):
"""
Encode the given batch of inputs. This method accept both raw text sequences
as well as already pre-tokenized sequences.
as well as already pre-tokenized sequences. The reason we use `PySequence` is
because it allows type checking with zero-cost (according to PyO3) as we don't
have to convert to check.
Example:
Here are some examples of the inputs that are accepted::

54
bindings/python/src/tokenizer.rs
View File

@ -995,7 +995,9 @@ impl PyTokenizer {
}
/// Encode the given batch of inputs. This method accept both raw text sequences
/// as well as already pre-tokenized sequences.
/// as well as already pre-tokenized sequences. The reason we use `PySequence` is
/// because it allows type checking with zero-cost (according to PyO3) as we don't
/// have to convert to check.
///
/// Example:
/// Here are some examples of the inputs that are accepted::
@ -1030,25 +1032,24 @@ impl PyTokenizer {
fn encode_batch(
&self,
py: Python<'_>,
input: Bound<'_, PyList>,
input: Bound<'_, PySequence>,
is_pretokenized: bool,
add_special_tokens: bool,
) -> PyResult<Vec<PyEncoding>> {
let input: Vec<tk::EncodeInput> = input
.into_iter()
.map(|o| {
let input: tk::EncodeInput = if is_pretokenized {
o.extract::<PreTokenizedEncodeInput>()?.into()
} else {
o.extract::<TextEncodeInput>()?.into()
};
Ok(input)
})
.collect::<PyResult<Vec<tk::EncodeInput>>>()?;
let mut items = Vec::<tk::EncodeInput>::with_capacity(input.len()?);
for i in 0..input.len()? {
let item = input.get_item(i)?;
let item: tk::EncodeInput = if is_pretokenized {
item.extract::<PreTokenizedEncodeInput>()?.into()
} else {
item.extract::<TextEncodeInput>()?.into()
};
items.push(item);
}
py.allow_threads(|| {
ToPyResult(
self.tokenizer
.encode_batch_char_offsets(input, add_special_tokens)
.encode_batch_char_offsets(items, add_special_tokens)
.map(|encodings| encodings.into_iter().map(|e| e.into()).collect()),
)
.into()
@ -1091,25 +1092,24 @@ impl PyTokenizer {
fn encode_batch_fast(
&self,
py: Python<'_>,
input: Bound<'_, PyList>,
input: Bound<'_, PySequence>,
is_pretokenized: bool,
add_special_tokens: bool,
) -> PyResult<Vec<PyEncoding>> {
let input: Vec<tk::EncodeInput> = input
.into_iter()
.map(|o| {
let input: tk::EncodeInput = if is_pretokenized {
o.extract::<PreTokenizedEncodeInput>()?.into()
} else {
o.extract::<TextEncodeInput>()?.into()
};
Ok(input)
})
.collect::<PyResult<Vec<tk::EncodeInput>>>()?;
let mut items = Vec::<tk::EncodeInput>::with_capacity(input.len()?);
for i in 0..input.len()? {
let item = input.get_item(i)?;
let item: tk::EncodeInput = if is_pretokenized {
item.extract::<PreTokenizedEncodeInput>()?.into()
} else {
item.extract::<TextEncodeInput>()?.into()
};
items.push(item);
}
py.allow_threads(|| {
ToPyResult(
self.tokenizer
.encode_batch_fast(input, add_special_tokens)
.encode_batch_fast(items, add_special_tokens)
.map(|encodings| encodings.into_iter().map(|e| e.into()).collect()),
)
.into()