llama.cpp
model : avoid ggml_cont_3d for fused QKV weights
#15662

Merged

model : avoid ggml_cont_3d for fused QKV weights #15662

ggerganov merged 7 commits into master from gg/model-avoid-cont3d

ggerganov marked this pull request as ready for review 60 days ago

model : avoid ggml_cont_3d for fused QKV weights

bb1202b2

kv-cache : make cpy_k and cpy_v implementation more readable

85a5ea36

cont : add comments

3dec397b

ggerganov force pushed from f15d515e to 3dec397b 60 days ago

cont : minor fix [no ci]

c62c354f

cont : one more fix

1efa9e8a

cont : clarity

d6be191d

CISC approved these changes on 2025-09-08

kv-cache : require contiguous heads of k_cur and v_cur

60d6e7c6

ggerganov merged cf0e3ba1 into master 60 days ago

ggerganov deleted the gg/model-avoid-cont3d branch 60 days ago

Reviewers

CISC

Assignees

No one assigned

Labels

None yet

Milestone

No milestone