Skip to content

Commit

Permalink
Enable dropout = 0.0 as an equivalent to none in BPE (#1550)
Browse files Browse the repository at this point in the history
* enable dropout = 0.0

* typo

* lint

* formatter

* enable dropout = 0.0

* formatter
  • Loading branch information
mcognetta authored Jun 24, 2024
1 parent 9441f7e commit fdd26ba
Show file tree
Hide file tree
Showing 4 changed files with 35 additions and 4 deletions.
4 changes: 4 additions & 0 deletions bindings/python/tests/bindings/test_models.py
Original file line number Diff line number Diff line change
Expand Up @@ -69,6 +69,10 @@ def test_can_modify(self):
model.byte_fallback = True
assert model.byte_fallback == True

def test_dropout_zero(self):
model = BPE(dropout=0.0)
assert model.dropout == 0.0


class TestWordPiece:
def test_instantiate(self, bert_files):
Expand Down
2 changes: 1 addition & 1 deletion tokenizers/src/models/bpe/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,7 @@ pub enum Error {
#[error("Unk token `{0}` not found in the vocabulary")]
UnkTokenOutOfVocabulary(String),
/// Dropout not between 0 and 1.
#[error("Dropout should be between 0 and 1")]
#[error("Dropout should be between 0 and 1, inclusive")]
InvalidDropout,
}

Expand Down
18 changes: 15 additions & 3 deletions tokenizers/src/models/bpe/model.rs
Original file line number Diff line number Diff line change
Expand Up @@ -136,7 +136,7 @@ impl BpeBuilder {
pub fn build(mut self) -> Result<BPE> {
// Validate dropout.
if let Some(p) = self.config.dropout {
if p <= 0.0 || p > 1.0 {
if !(0.0..=1.0).contains(&p) {
return Err(Error::InvalidDropout.into());
}
}
Expand Down Expand Up @@ -214,7 +214,7 @@ pub struct BPE {
pub(crate) merges: MergeMap,
/// Contains the cache for optimizing the encoding step.
cache: Option<Cache<String, Word>>,
/// Dropout probability for merges. 0 = no dropout is the default. At 1.0, tokenization will
/// Dropout probability for merges. 0.0 = no dropout is the default. At 1.0, tokenization will
/// perform no merges, so the result will just be characters.
pub dropout: Option<f32>,
/// The unknown token to be used when we encounter an unknown char
Expand Down Expand Up @@ -493,7 +493,7 @@ impl Model for BPE {
return Ok(vec![]);
}

if self.dropout.is_none() {
if self.dropout.is_none() || self.dropout == Some(0.0) {
self.tokenize_with_cache(sequence)
} else {
let word = self.merge_word(sequence)?;
Expand Down Expand Up @@ -685,6 +685,11 @@ mod tests {
let tokens = bpe.tokenize("unrelated").unwrap();
assert_eq!(tokens, vec![Token::new(15u32, "unrelated".into(), (0, 9))]);

// With dropout = 0.0 (equivalent to dropout == none)
bpe.dropout = Some(0.0);
let tokens = bpe.tokenize("unrelated").unwrap();
assert_eq!(tokens, vec![Token::new(15u32, "unrelated".into(), (0, 9))]);

// Now set dropout to 1.0. Result should be no merges performed.
bpe.dropout = Some(1.0);
let tokens = bpe.tokenize("unrelated").unwrap();
Expand Down Expand Up @@ -739,6 +744,13 @@ mod tests {
assert_eq!(bpe.vocab.get("ab").unwrap(), &3u32);
}

#[test]
// Ensure BPEBuilder with dropout = 0.0 doesn't error
fn test_bpe_with_dropout_0() {
let bpe = BPE::builder().dropout(0.0).build().unwrap();
assert_eq!(bpe.dropout, Some(0.0));
}

#[test]
// Ensure `BPE::from_file` works as expected.
fn test_bpe_with_continuing_subword_prefix() {
Expand Down
15 changes: 15 additions & 0 deletions tokenizers/tests/serialization.rs
Original file line number Diff line number Diff line change
Expand Up @@ -229,6 +229,21 @@ fn tokenizer() {
assert_eq!(serde_json::to_string(&de).unwrap(), ser);
}

#[test]
fn bpe_with_dropout_serde() {
let mut bpe = BPE::default();
bpe.dropout = Some(0.1);
let ser = serde_json::to_string(&bpe).unwrap();
let de = serde_json::from_str(&ser).unwrap();
assert_eq!(bpe, de);

// set dropout to 0.0 (which is analogous to None) and reserialize
bpe.dropout = Some(0.0);
let ser = serde_json::to_string(&bpe).unwrap();
let de = serde_json::from_str(&ser).unwrap();
assert_eq!(bpe, de);
}

#[test]
fn test_deserialize_long_file() {
let _tokenizer = Tokenizer::from_file("data/albert-base-v1-tokenizer.json").unwrap();
Expand Down

0 comments on commit fdd26ba

Please sign in to comment.