Repository navigation
Fix qmv_impl for small N - #3096
Merged
Merged
Conversation
manuelcandales
changed the title
Member
|
I would imagine we have the same bug in Do you mind fixing it there as well and updating the test to include another quantization mode like |
awni
reviewed
Feb 4, 2026
| self.assertEqual(y_q.shape, y_hat.shape) | ||
| self.assertLess((y_q - y_hat).abs().max(), 1e-3) | ||
|
|
||
| def test_small_non_multiples(self): |
Member
There was a problem hiding this comment.
Nit: could you change this function name to tests_qmv_small_non_multiple so it's a bit more descriptive.
Member
|
Good catch! We can merge this as soon as comments are addressed! |
manuelcandales
added a commit
to pytorch/executorch
that referenced
this pull request
Feb 5, 2026
Fixes two bugs in `qmv_impl` (also submitted [PR 3096](ml-explore/mlx#3096) to MLX) * Fix 1: Changed `qdot` to `qdot_safe` when handling remaining elements in the `N < 8` case Notice that in `qmv_impl`, there are two code paths based on whether `N < 8`: - When `N >= 8`: Correctly uses `qdot_safe` for remaining elements - When `N < 8`: Incorrectly used `qdot` instead of `qdot_safe` for remaining elements (fixed in this PR) * Fix 2: Changed `in_vec_size_g` from floor division to ceiling division to correctly handle cases where `K` is not a multiple of group size
awni
force-pushed
the
manuel/fix-qmv-impl
branch
from
February 5, 2026 20:44
fe9d0d6 to
5831631
Compare
awni
force-pushed
the
manuel/fix-qmv-impl
branch
from
February 5, 2026 23:51
056b769 to
d7b3bd4
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Proposed changes
Fixes two bugs in
qmv_impl:Fix 1: Changed
qdottoqdot_safewhen handling remaining elements in theN < 8caseNotice that in
qmv_impl, there are two code paths based on whetherN < 8:N >= 8: Correctly usesqdot_safefor remaining elementsN < 8: Incorrectly usedqdotinstead ofqdot_safefor remaining elements (fixed in this PR)Fix 2: Changedin_vec_size_gfrom floor division to ceiling division to correctly handle cases whereKis not a multiple of group sizeEDIT: I have reverted Fix 2, since it seems that MLX never allows K to be non-multiple of group_size
Checklist
Put an
xin the boxes that apply.pre-commit run --all-filesto format my code / installed pre-commit prior to committing changes