Skip to content

fix: guard quantiseAndSend against zero-length sequence and hidden dim #78

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

Merged
devlux76 merged 2 commits into from
Mar 21, 2026
Merged

fix: guard quantiseAndSend against zero-length sequence and hidden dim #78

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e6aedaf
Initial plan
Copilot Mar 21, 2026
104fc66
fix: add seqLen < 1 and hiddenDim < 1 guards to quantiseAndSend
Copilot Mar 21, 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
8 changes: 8 additions & 0 deletions src/worker.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,6 +281,14 @@ async function quantiseAndSend(
hiddenDim: number,
dtype: EmbeddingMsg['dtype'],
): Promise<void> {
if (seqLen < 1) {
workerLog('warn', 'quantiseAndSend: seqLen < 1; skipping Q² quantisation', { seqLen });
return;
}
if (hiddenDim < 1) {
workerLog('warn', 'quantiseAndSend: hiddenDim < 1; skipping Q² quantisation', { hiddenDim });
return;
}
Comment on lines +284 to +291

Copilot AI Mar 21, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The new guards prevent the seqLen=0/hiddenDim=0 trap, but quantiseAndSend still doesn’t validate other documented kernel preconditions (e.g., n must be a power of 2 and ≤ 16,384, and the buffer must contain at least the last-token row). Without these checks it’s still possible to call into the WASM kernel with unsupported dimensions or an undersized buffer, which can lead to traps or incorrect fingerprints. Consider extending the early validation to enforce the kernel’s n constraints (and, if applicable, the expected byteLength for the given dtype) before copying/calling into WASM.

Copilot uses AI. Check for mistakes.
Comment on lines +284 to +291

Copilot AI Mar 21, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This change introduces new early-return behavior for invalid seqLen/hiddenDim, but there’s no unit test asserting that the Q² kernel is not invoked in these cases. Since the repo already has worker unit tests, consider adding a test that triggers the embedding extraction path with seqLen=0 and/or hiddenDim=0 and verifies getKernel/quantise are not called (and that the worker continues without error).

Copilot uses AI. Check for mistakes.
const n = hiddenDim;
const dtypeId = DTYPE_TO_Q2[dtype] ?? Q2_DTYPE_FP32;
try {
Expand Down
Loading