llama.cpp
Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support
#16900

Merged

Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support #16900

0cc4m merged 15 commits into master from 0cc4m/vulkan-mmq-dp4a-vec-k-quants

github-actions added Vulkan

github-actions added ggml

0cc4m force pushed from d5192bf4 to d2f8f00f 85 days ago

0cc4m marked this pull request as ready for review 85 days ago

0cc4m requested a review from

jeffbolznv 85 days ago

jeffbolznv commented on 2025-11-01

0cc4m force pushed from b153aac3 to 1b78909c 79 days ago

0cc4m force pushed from 1b78909c to 937f9925 71 days ago

0cc4m force pushed from b99726ce to 3c22e380 71 days ago

0cc4m force pushed from 3c22e380 to e086733b 67 days ago

0cc4m force pushed from e086733b to e69d645a 64 days ago

0cc4m marked this pull request as draft 64 days ago

0cc4m force pushed to 9cbe4f87 64 days ago

github-actions added testing

vulkan: split mul_mmq_funcs for mul_mat_vecq use

f7a638f2

add mxfp4 mmvq

211bcd49

add q2_k mmvq

9ba12585

add q3_k mmvq

9eeb42fc

add q4_k and q5_k mmvq

741bf821

add q6_k mmvq

7a8b853a

handle 4x4 quants per mmvq thread

ef0060a9

enable MUL_MAT_ID mmvq support

593e94fb

enable subgroup optimizations for mul_mat_vec_id shaders

bb3bcaab

device tuning

4512c55b

request prealloc_y sync after quantization

dd54d395

fix indentation

b119ac78

fix llvmpipe test failures

f8e32881

fix mul_mat_id mmvq condition

ad5127d9

0cc4m force pushed from 9cbe4f87 to ad5127d9 60 days ago

0cc4m marked this pull request as ready for review 60 days ago

0cc4m requested a review from

jeffbolznv 59 days ago

jeffbolznv commented on 2025-11-28

fix unused variable warning

6cb0923b

jeffbolznv approved these changes on 2025-11-29

0cc4m merged 47a268ea into master 58 days ago

0cc4m deleted the 0cc4m/vulkan-mmq-dp4a-vec-k-quants branch 58 days ago

Reviewers

jeffbolznv

Assignees

No one assigned

Labels

testing Vulkan ggml

Milestone

No milestone

llama.cpp Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support #16900 Merged

Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support #16900

llama.cpp
Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support
#16900

Merged