llama.cpp
vulkan: do not enable MMVQ for speculative-decode steps on AMD
#25666

Open

vulkan: do not enable MMVQ for speculative-decode steps on AMD #25666

SnoopsDev wants to merge 2 commits into ggml-org:master from SnoopsDev:vulkan-mmvq-speculative-decode

vulkan: don't use MMVQ for k-quants at small k on AMD

fe8dbd2a

SnoopsDev requested a review 4 days ago

github-actions added Vulkan

github-actions added ggml

jeffbolznv commented on 2026-07-14

vulkan: treat a speculative-decode step as decode, not as a batch

e0649d3e

SnoopsDev force pushed from f347112e to e0649d3e 4 days ago

Reviewers

jeffbolznv

Assignees

No one assigned

Labels

Vulkan ggml

Milestone

No milestone