llama.cpp
llama : fix unexpected graph reallocation in the k-pool models
#29958
Merged
Merged
llama : fix unexpected graph reallocation in the k-pool models #29958
ggerganov
merged
8 commits into
master
from
gg/sparse-fix-realloc
llama : fix unexpected graph reallocation in the k-pool models
2220411e
llama : drop the unused k-pool cache_safe graph API
46d2af84
tests : add a shared-seq graph reserve regression test
45e8cb9b
cont : add TODOs
d69cf37e
cont : fix comment
30e9f978
ggerganov
commented
on 2026-10-04
am17an
approved these changes
on 2026-10-05
cuda: match the moe weighted reduction on empty ubatches
bcc7ebc6
ServeurpersoCom
approved these changes
on 2026-10-05
tests: build the rollback test only where internal symbols link
20ace9f9
ggerganov
commented
on 2026-10-05
tests: skip archs by name in the shared-seq reserve test
2211ab6f
Reviewers
Assignees
No one assigned
Labels
model
testing
ggml
CUDA
Milestone
No milestone
Login to write a write a comment.
Login via GitHub