Skip to content

feat(pipeline+bindings): Support encoding pairs of sequences #2237

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We鈥檒l occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
SBrandeis wants to merge 1 commit into
base: feat/pipeline-python-api
Choose a base branch
from
Draft

feat(pipeline+bindings): Support encoding pairs of sequences #2237

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 21 additions & 6 deletions bindings/python/src/tokenizer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,6 +42,16 @@ fn poisoned<G>(_: std::sync::PoisonError<G>) -> PyErr {
PyRuntimeError::new_err("tokenizer lock poisoned")
}

/// The input sequence(s) to [`encode`](`PyTokenizer::encode`)
/// Can be either a single str or a pair of (str, str)
#[derive(FromPyObject)]
enum InputSequence {
#[pyo3(annotation = "str")]
Single(PyBackedStr),
#[pyo3(annotation = "(str, str)")]
Pair(PyBackedStr, PyBackedStr),
}

/// A tokenizer: a model plus its optional normalizer and pre-tokenizer.
///
/// Create one from a model (`Tokenizer(models.BPE())`), a file
Expand Down Expand Up @@ -173,7 +183,7 @@ impl PyTokenizer {
fn run_encode(
&self,
py: Python<'_>,
text: &str,
text: &InputSequence,
add_special_tokens: bool,
) -> PyResult<Vec<u32>> {
self.inner.with(py, |lock| {
Expand All @@ -197,7 +207,7 @@ impl PyTokenizer {
fn run_encode_batch(
&self,
py: Python<'_>,
texts: &[PyBackedStr],
texts: &[InputSequence],
add_special_tokens: bool,
) -> PyResult<Vec<Vec<u32>>> {
self.inner.with(py, |lock| {
Expand Down Expand Up @@ -285,7 +295,7 @@ fn get_or_compile(lock: &Detached<'_, Inner>) -> PyResult<Arc<PipelineTokenizer>

fn encode_one(
pipe: &PipelineTokenizer,
text: &str,
text: InputSequence,
pre_tokens: &mut Vec<Span>,
add_special_tokens: bool,
scratch: &mut PipelineModelScratch,
Expand All @@ -308,7 +318,7 @@ fn encode_one(
/// and the batch is worth splitting; the caller has already released the GIL.
fn encode_batch_core(
pipe: &PipelineTokenizer,
texts: &[PyBackedStr],
texts: &[InputSequence],
add_special_tokens: bool,
) -> PyResult<Vec<Vec<u32>>> {
if get_parallelism() && texts.len() > 1 {
Expand Down Expand Up @@ -428,7 +438,7 @@ impl PyTokenizer {
fn encode_batch_ids<'py>(
&self,
py: Python<'py>,
texts: Vec<PyBackedStr>,
texts: Vec<InputSequence>,
add_special_tokens: bool,
) -> PyResult<Bound<'py, PyList>> {
let batches = self.run_encode_batch(py, &texts, add_special_tokens)?;
Expand Down Expand Up @@ -479,7 +489,7 @@ impl PyTokenizer {
#[pyo3(signature = (texts, *, add_special_tokens = true) -> "EncodingBatch")]
fn encode_batch(
slf: &Bound<'_, Self>,
texts: Vec<PyBackedStr>,
texts: Vec<InputSequence>,
add_special_tokens: bool,
) -> PyResult<PyEncodingBatch> {
let rows = slf
Expand Down Expand Up @@ -764,3 +774,8 @@ impl Iterator for BufferedPyIterator {
self.buffer.pop_front()
}
}


struct PyEncodeInput {

}
Loading
Loading