build: Reuse the extensions for a Python-only patch series

git am refreshes every file timestamp, so a Python-only patch series
forced a full extension rebuild and ninja saw everything as changed.
Diff csrc/ and the build files between the recorded tree and HEAD, and
rebuild only when a C++/CUDA source moved; otherwise the shipped .so
stays.

Update the expected tree to the 106-patch series
(c9c526755364d44d8d9923d9acc9d6d1a3d58718).

Add ipc: host, SYS_PTRACE, and seccomp=unconfined to the container so
a profiler or debugger can attach.

Raise --block-size from 2048 to 4096 and drop
--safetensors-load-strategy prefetch.

Co-Authored-By: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
2026-10-17 23:29:02 +00:00
co-authored by Hinata
parent 50dabc9989
commit fe3cd9247b
3 changed files with 18 additions and 4 deletions
+5
View File
@@ -19,7 +19,12 @@ services:
count: all
capabilities: [gpu]
command: ["${MODEL_MOUNT_PATH:-/models/Qwen3.8-Flash-Next}"]
ipc: host
shm_size: 16g
cap_add:
- SYS_PTRACE
security_opt:
- seccomp=unconfined
ulimits:
memlock: -1
stack: 67108864