Skip to content
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’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Bpe clones #1707

Open
wants to merge 6 commits into
base: main
Choose a base branch
from
+49 −115
Open

Bpe clones #1707

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5947e34
Dont iterate over chars to determine length
sftse Dec 24, 2024
2fabb1f
Lose some unnecessary allocations
sftse Dec 29, 2024
783fe7a
fmt
sftse Dec 29, 2024
9659474
Small refactors
sftse Dec 24, 2024
54b1b78
Remove needless unwrap
sftse Dec 24, 2024
7ba6512
Subtyping will autocoerce lifetimes by itself
sftse Dec 28, 2024
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
28 changes: 12 additions & 16 deletions tokenizers/src/decoders/wordpiece.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,23 +45,19 @@ pub fn cleanup(dirty_input: &str) -> String {

impl Decoder for WordPiece {
fn decode_chain(&self, mut tokens: Vec<String>) -> Result<Vec<String>> {
tokens
.iter_mut()
.enumerate()
.map(|(i, token)| {
if i != 0 {
if token.starts_with(&self.prefix) {
*token = token.replacen(&self.prefix, "", 1);
} else {
*token = format!(" {token}");
}
for (i, token) in tokens.iter_mut().enumerate() {
if i != 0 {
if let Some(tk) = token.strip_prefix(&self.prefix) {
*token = tk.to_string();
} else {
*token = format!(" {token}");
}
if self.cleanup {
*token = cleanup(token);
}
Ok(token.to_string())
})
.collect::<Result<_>>()
}
if self.cleanup {
*token = cleanup(token);
}
}
Ok(tokens)
}
}

Expand Down
4 changes: 2 additions & 2 deletions tokenizers/src/models/bpe/model.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -460,7 +460,7 @@ impl BPE {
Ok(word)
}

fn word_to_tokens<'a, 'b: 'a>(&'a self, word: &'b Word) -> impl Iterator<Item = Token> + 'a {
fn word_to_tokens<'a>(&'a self, word: &'a Word) -> impl Iterator<Item = Token> + 'a {
word.get_chars_iter()
.zip(word.get_offsets_iter())
.map(move |(id, offsets)| Token::new(id, self.vocab_r[&id].clone(), offsets))
Expand All @@ -471,7 +471,7 @@ impl BPE {
if let Some(id) = self.vocab.get(sequence) {
return Ok(vec![Token::new(
*id,
sequence.to_string().clone(),
sequence.to_string(),
(0, sequence.len()),
)]);
}
Expand Down
93 changes: 21 additions & 72 deletions tokenizers/src/models/bpe/trainer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -271,19 +271,13 @@ impl BpeTrainer {
let mut alphabet: HashMap<char, usize> = HashMap::new();
for (word, count) in wc {
for c in word.chars() {
alphabet
.entry(c)
.and_modify(|cnt| *cnt += *count as usize)
.or_insert(*count as usize);
*alphabet.entry(c).or_default() += *count as usize;
}
}

// Also include anything from the provided initial alphabet
for c in &self.initial_alphabet {
alphabet
.entry(*c)
.and_modify(|cnt| *cnt = usize::MAX)
.or_insert(usize::MAX);
*alphabet.entry(*c).or_default() = usize::MAX;
}

let mut kept = alphabet.iter().collect::<Vec<_>>();
Expand All @@ -293,13 +287,7 @@ impl BpeTrainer {
// will be removed
let to_remove = self
.limit_alphabet
.map(|limit| {
if alphabet.len() > limit {
alphabet.len() - limit
} else {
0
}
})
.map(|limit| alphabet.len().saturating_sub(limit))
.unwrap_or(0);

// Remove the unwanted chars
Expand All @@ -309,7 +297,7 @@ impl BpeTrainer {
}

// Keep the initial alphabet (sorted for determinism)
kept.sort_unstable_by_key(|k| (*k.0) as u32);
kept.sort_unstable_by_key(|k| *k.0 as u32);
kept.into_iter().for_each(|(c, _)| {
let s = c.to_string();
if !w2id.contains_key(&s) {
Expand Down Expand Up @@ -342,13 +330,13 @@ impl BpeTrainer {
// Add the `continuing_subword_prefix` if relevant
if !is_first {
if let Some(prefix) = &self.continuing_subword_prefix {
s = format!("{prefix}{s}");
s.insert_str(0, prefix);
}
}
// Add the `end_of_word_suffix` if relevant
if is_last {
if let Some(suffix) = &self.end_of_word_suffix {
s = format!("{s}{suffix}");
s.push_str(suffix);
}
}

Expand Down Expand Up @@ -387,23 +375,9 @@ impl BpeTrainer {
let cur_pair: Pair = (window[0], window[1]);

// Initialize pair_counts and where_to_update for this pair if we just saw it
if !pair_counts.contains_key(&cur_pair) {
pair_counts.insert(cur_pair, 0);
}

// Then update counts
let count = counts[i];
where_to_update
.entry(cur_pair)
.and_modify(|h| {
h.insert(i);
})
.or_insert_with(|| {
let mut h = HashSet::new();
h.insert(i);
h
});
*pair_counts.get_mut(&cur_pair).unwrap() += count as i32;
*pair_counts.entry(cur_pair).or_default() += counts[i] as i32;
where_to_update.entry(cur_pair).or_default().insert(i);
}

if let Some(p) = &p {
Expand All @@ -416,13 +390,10 @@ impl BpeTrainer {
|| (HashMap::new(), HashMap::new()),
|(mut pair_counts, mut where_to_update), (pc, wtu)| {
for (k, v) in pc {
pair_counts.entry(k).and_modify(|c| *c += v).or_insert(v);
*pair_counts.entry(k).or_default() += v;
}
for (k, v) in wtu {
where_to_update
.entry(k)
.and_modify(|set| *set = set.union(&v).copied().collect())
.or_insert(v);
where_to_update.entry(k).or_default().extend(v);
}
(pair_counts, where_to_update)
},
Expand Down Expand Up @@ -488,11 +459,10 @@ impl BpeTrainer {
break;
}

if queue.is_empty() {
let Some(mut top) = queue.pop() else {
break;
}
};

let mut top = queue.pop().unwrap();
if top.count != pair_counts[&top.pair] as u64 {
top.count = pair_counts[&top.pair] as u64;
queue.push(top);
Expand All @@ -504,13 +474,12 @@ impl BpeTrainer {
}

let part_a = &id_to_word[top.pair.0 as usize];
let mut part_b = id_to_word[top.pair.1 as usize].to_owned();
let mut part_b = id_to_word[top.pair.1 as usize].as_str();

// Build new token
if let Some(prefix) = &self.continuing_subword_prefix {
if part_b.starts_with(prefix) {
let prefix_byte_len = prefix.chars().map(|c| c.len_utf8()).sum();
part_b = part_b[prefix_byte_len..].to_string();
if let Some(rest) = part_b.strip_prefix(prefix) {
part_b = rest;
}
}
let new_token = format!("{part_a}{part_b}");
Expand Down Expand Up @@ -566,21 +535,9 @@ impl BpeTrainer {
// Introduce new formed pairs
for ((pair, change), iw) in changes {
let count = change * counts[iw] as i32;
pair_counts
.entry(pair)
.and_modify(|c| *c += count)
.or_insert(count);
*pair_counts.entry(pair).or_default() += count;
if change > 0 {
where_to_update
.entry(pair)
.and_modify(|h| {
h.insert(iw);
})
.or_insert_with(|| {
let mut h = HashSet::new();
h.insert(iw);
h
});
where_to_update.entry(pair).or_default().insert(iw);
}
}
where_to_update.drain().for_each(|(pair, pos)| {
Expand Down Expand Up @@ -613,16 +570,8 @@ impl BpeTrainer {
.map(|(i, (pair, new_token_id))| (pair, (i as u32, new_token_id)))
.collect();

if let Some(prefix) = &self.continuing_subword_prefix {
model.continuing_subword_prefix = Some(prefix.to_owned());
} else {
model.continuing_subword_prefix = None;
}
if let Some(suffix) = &self.end_of_word_suffix {
model.end_of_word_suffix = Some(suffix.to_owned());
} else {
model.end_of_word_suffix = None;
}
model.continuing_subword_prefix = self.continuing_subword_prefix.clone();
model.end_of_word_suffix = self.end_of_word_suffix.clone();

Ok(self.special_tokens.clone())
}
Expand Down Expand Up @@ -653,7 +602,7 @@ impl Trainer for BpeTrainer {
let words = process(sequence.as_ref())?;
let mut map = HashMap::new();
for word in words {
map.entry(word).and_modify(|c| *c += 1).or_insert(1);
*map.entry(word).or_default() += 1;
}
Ok(map)
})
Expand All @@ -662,7 +611,7 @@ impl Trainer for BpeTrainer {
|acc, ws| {
let mut acc = acc?;
for (k, v) in ws? {
acc.entry(k).and_modify(|c| *c += v).or_insert(v);
*acc.entry(k).or_default() += v;
}
Ok(acc)
},
Expand Down
18 changes: 5 additions & 13 deletions tokenizers/src/models/unigram/model.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -115,8 +115,7 @@ impl Unigram {
let mut min_score = f64::INFINITY;
for (id, (token, score)) in vocab.iter().enumerate() {
token_to_ids.insert(token.to_string(), id as u32);
let bytes: Vec<u8> = token.bytes().collect();
builder.push(&bytes);
builder.push(token.as_bytes());
if score < &min_score {
min_score = *score;
}
Expand Down Expand Up @@ -308,22 +307,15 @@ impl Unigram {
while ends_at > 0 {
let node = &best_path_ends_at[ends_at];
let starts_at = node.starts_at.unwrap();
if self.fuse_unk
&& self.unk_id.is_some()
&& node.id == self.unk_id.ok_or(UnigramError::MissingUnkId)?
{
token.push(
String::from_utf8(sentence[starts_at..ends_at].as_bytes().to_vec()).unwrap(),
);
if self.fuse_unk && Some(node.id) == self.unk_id {
token.push(sentence[starts_at..ends_at].to_string());
} else {
if !token.is_empty() {
token.reverse();
results.push(token.concat());
token = vec![];
}
results.push(
String::from_utf8(sentence[starts_at..ends_at].as_bytes().to_vec()).unwrap(),
);
results.push(sentence[starts_at..ends_at].to_string());
}
ends_at = starts_at;
}
Expand All @@ -350,7 +342,7 @@ impl Unigram {
results.push(token);
token = String::new();
}
results.push(item.to_string());
results.push(item);
}
}
if !token.is_empty() {
Expand Down
6 changes: 3 additions & 3 deletions tokenizers/src/models/unigram/trainer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -428,7 +428,7 @@ impl UnigramTrainer {
new_pieces.push(pieces[id].clone());
}

new_pieces.to_vec()
new_pieces
}

/// Update the progress bar with the new provided length and message
Expand Down Expand Up @@ -637,7 +637,7 @@ impl Trainer for UnigramTrainer {
let words = process(sequence.as_ref())?;
let mut map = HashMap::new();
for word in words {
map.entry(word).and_modify(|c| *c += 1).or_insert(1);
*map.entry(word).or_default() += 1;
}
Ok(map)
})
Expand All @@ -646,7 +646,7 @@ impl Trainer for UnigramTrainer {
|acc, ws| {
let mut acc = acc?;
for (k, v) in ws? {
acc.entry(k).and_modify(|c| *c += v).or_insert(v);
*acc.entry(k).or_default() += v;
}
Ok(acc)
},
Expand Down
4 changes: 2 additions & 2 deletions tokenizers/src/models/wordlevel/trainer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@ impl Trainer for WordLevelTrainer {
let words = process(sequence.as_ref())?;
let mut map = HashMap::new();
for word in words {
map.entry(word).and_modify(|c| *c += 1).or_insert(1);
*map.entry(word).or_default() += 1;
}
Ok(map)
})
Expand All @@ -115,7 +115,7 @@ impl Trainer for WordLevelTrainer {
|acc, ws| {
let mut acc = acc?;
for (k, v) in ws? {
acc.entry(k).and_modify(|c| *c += v).or_insert(v);
*acc.entry(k).or_default() += v;
}
Ok(acc)
},
Expand Down
11 changes: 4 additions & 7 deletions tokenizers/src/tokenizer/added_vocabulary.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -297,15 +297,12 @@ impl AddedVocabulary {
)
};
// Make sure we modify the previous entry
self.added_tokens_map
*self
.added_tokens_map
.entry(token.content.clone())
.and_modify(|old_id| *old_id = new_id)
.or_insert_with(|| new_id);
.or_default() = new_id;
// Update the current revert operation
self.added_tokens_map_r
.entry(new_id)
.and_modify(|t| *t = token.clone())
.or_insert_with(|| token.clone());
*self.added_tokens_map_r.entry(new_id).or_default() = token.clone();
// Make sure to remove previous entry (if the token gets a new id)

// Finally add the token to the classic set if special
Expand Down