llama.cpp
metal : few-row MMA mat-mul and batched copies for speculative decoding
#29869

Merged

metal : few-row MMA mat-mul and batched copies for speculative decoding #29869

ggerganov merged 10 commits into ggml-org:master from pratiknarola-t:metal-few-row-mma

metal : few-row MMA mat-mul and batched copies for speculative decoding

cc470b64

github-actions added testing

github-actions added ggml

github-actions added Apple Metal

ggerganov assigned

ggerganov 8 days ago

ggerganov commented on 2026-10-02

metal : remove the CPY_BATCH fusion and the memory range changes

e0356f6e

Merge remote-tracking branch 'origin/master' into metal-few-row-mma

ac369808

ggerganov commented on 2026-10-04

cont : clean-up

5a993475

cont : drop has_tensor gate

8c084b0b

cont : clean-up operand/residual logic

3af8f157

cont : drop Q4_0 ne11=2 special-case

17acbc52

cont : add kernels/mul_mv_mma.metal

9cf70aca

cont : consolidate mma pipeline selection logic

63ca857f

cont : decouple fusion logic from device props

da88932a

ggerganov marked this pull request as ready for review 6 days ago

ggerganov requested a review 6 days ago

ggerganov approved these changes on 2026-10-05

ggerganov merged a3a1c474 into master 5 days ago

ggerganov added highlight

Reviewers

ggerganov

Assignees

ggerganov

Labels

testing ggml Apple Metal highlight

Milestone

No milestone