llama.cpp
llama: fix k-pool scatter data race on shared sequences
#29994

Merged

llama: fix k-pool scatter data race on shared sequences #29994

ServeurpersoCom merged 3 commits into ggml-org:master from ServeurpersoCom:kpool-shared-rep-tsan

llama: re-pool each shared k-pool rep once

5a2e623c

llama: assert whole-sequence seq_cp in the hybrid idx memory

9dfbcd03

ServeurpersoCom requested a review from

ggerganov 30 days ago

ServeurpersoCom requested a review from

CISC 30 days ago

ServeurpersoCom requested a review from

am17an 30 days ago

am17an commented on 2026-10-05

llama: drop the k-pool cache_safe mode

42b3cf56

ServeurpersoCom requested a review from

am17an 29 days ago

github-actions added model

am17an approved these changes on 2026-10-06

ggerganov approved these changes on 2026-10-06

ServeurpersoCom merged 43fe9c64 into master 29 days ago

Reviewers

ggerganov

am17an

CISC

Assignees

No one assigned

Labels

model

Milestone

No milestone