Repository navigation
Conversation
dhiltgen
force-pushed
the
qmm-small-rows
branch
from
October 16, 2026 15:57
1e882b4 to
8d6fa26
Compare
zcbenz
added
the
await verification
zcbenz
force-pushed
the
qmm-small-rows
branch
from
October 19, 2026 03:42
8d6fa26 to
542c7bf
Compare
dhiltgen
added a commit
to dhiltgen/ollama
that referenced
this pull request
Oct 20, 2026
Quantized matmul tiled for 6-16 rows, the shape of speculative verify, rebased onto ml-explore/mlx#4709. Applied to the pinned MLX until the pin includes it; drop this commit then.
This branch has not been deployed
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This change speeds up speculative and batched decoding fp quant models when generating 6-16 sequences. Main today either reads all weights once per row, or Split-K qmm pads to 32-row tiles and wastes most of its work. This optimization adds a qmv_mma (M1-M4) using 8x8 simdgroup matrices and qmv_nax, which read each block of weights once, unpack it and multiply the rows on the matrix units to reduce memory traffic. This achieves ~3x speedup on M1/M2 generation, and ~2x on newer systems.
mlx_lm.benchmark -p 512 -g 128 -b <batch>performance examples:M5 Max
(No observed prompt or batch 1 change)