Skip to content

Fix i32 overflow in BPE trainer pair_counts on large corpora #2057

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

Draft
petyb wants to merge 3 commits into
base: main
Choose a base branch
from
+3 −3
Draft

Fix i32 overflow in BPE trainer pair_counts on large corpora #2057

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ab9c790
change i32 to i64
petyb May 14, 2026
6f3fb49
add as i64 to left operand in multiplication
petyb May 14, 2026
8b5d2e2
Merge branch 'main' into fix/bpe-trainer-pair-count-i32-overflow
petyb May 14, 2026
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
6 changes: 3 additions & 3 deletions tokenizers/src/models/bpe/trainer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -416,7 +416,7 @@ impl BpeTrainer {
words: &[Word],
counts: &[u64],
p: &Option<ProgressBar>,
) -> (AHashMap<Pair, i32>, AHashMap<Pair, AHashSet<usize>>) {
) -> (AHashMap<Pair, i64>, AHashMap<Pair, AHashSet<usize>>) {
words
.maybe_par_iter()
.enumerate()
Expand All @@ -429,7 +429,7 @@ impl BpeTrainer {

// Initialize pair_counts and where_to_update for this pair if we just saw it
// Then update counts
*pair_counts.entry(cur_pair).or_default() += counts[i] as i32;
*pair_counts.entry(cur_pair).or_default() += counts[i] as i64;
where_to_update.entry(cur_pair).or_default().insert(i);
}

Expand Down Expand Up @@ -581,7 +581,7 @@ impl BpeTrainer {

// Introduce new formed pairs
for ((pair, change), iw) in changes {
let count = change * counts[iw] as i32;
let count = change as i64 * counts[iw] as i64;
*pair_counts.entry(pair).or_default() += count;
if change > 0 {
where_to_update.entry(pair).or_default().insert(iw);
Expand Down