llama.cpp
server: save and clear idle slots on new task (`--clear-idle`)
#20993
Merged
Merged
server: save and clear idle slots on new task (`--clear-idle`) #20993
ngxson
merged
11 commits into
ggml-org:master
from
yychyo:server-kv-keep-only-active
server: clear idle slots KV from VRAM (LLAMA_KV_KEEP_ONLY_ACTIVE)
c82cd357
yychyo
commented
on 2026-03-26
yychyo
commented
on 2026-03-26
yychyo
commented
on 2026-03-26
ggerganov
commented
on 2026-03-26
server: move idle slot KV clearing to slot release
2563b4ae
server: add --kv-clear-idle flag, enable by default
ced2476c
yychyo
changed the title
server: skip clearing last idle slot, clear on launch
f6c4a3d2
server: test --no-kv-clear-idle flag
698e9a3c
server: simplify on-release clearing loop
d658a62c
server: remove on-release KV clearing, keep launch-only
3d338c9d
cont : clean-up
09ed3c03
tests: update log strings after --clear-idle rename
645aebc4
ggerganov
approved these changes
on 2026-04-01
ggerganov
commented
on 2026-04-01
yychyo
changed the title
yychyo
changed the title
tests: use debug tags instead of log message matching
a22ba243
test: fix Windows CI by dropping temp log file unlink
5fc16981
ggerganov
approved these changes
on 2026-04-03
pwilkin
approved these changes
on 2026-04-03
ngxson
approved these changes
on 2026-04-03
Login to write a write a comment.
Login via GitHub