Pin unslothai/llama.cpp#210 (server-side preemption and exact concurrency) at afd3248b82 - #211

Open

danielhanchen wants to merge 14 commits into

pin/server-side-preemption-197

Commits on Sep 10, 2026

Repin unslothai/llama.cpp#210 at 1d5548978d: parking is off unless --preempt-ram is given
danielhanchen
committed
3bc1877
Repin unslothai/llama.cpp#210 at 15bcc6c7a8: the paged path reads src[5] only for its own op
danielhanchen
committed
cff7085
Repin #210 at 3e7d9530a: the CI fixes for the setenv, OpenVINO and park assertion legs
danielhanchen
committed
0585f09
Repin #210 at 2176de1ad: the test-exact-buft placement and the two server test timings
danielhanchen
committed
56476f2
Repin #210 at 0beed5117: the rotation test timeout, recompute replay stats, streamed responses preempt record and the condensed weight check comment
danielhanchen
committed
20a699a
Repin #210 at 7eb9cb780: the started-slot test lengths, the recurrent transfer refusal and the condensed test preamble
danielhanchen
committed
78c6d42
Repin #210 at d8daa8fc5: the failed-save park test decided by lengths
danielhanchen
committed
c391bb4
Repin #210 at 5b66acb4d: prompt counters reset when a recompute park restarts a prefill
danielhanchen
committed
aa2b5df
Repin #210 at 41abdfb23: batched notices carry index 0, tests wait for the log to settle
danielhanchen
committed
099562d
Repin #210 at ae1953905: overlapping requests by lengths, whole-sentence host buffer comment
danielhanchen
committed
aecf89a