feat(cmf): per-server O(1) attention override at serve time
Docker / publish (push) Canceled after 0s
Release / crates-io (push) Canceled after 0s
Release / release (macos-14, aarch64-apple-darwin) (push) Canceled after 0s
Release / release (ubuntu-latest, x86_64-unknown-linux-gnu) (push) Canceled after 0s
Release / release (windows-latest, x86_64-pc-windows-msvc) (push) Canceled after 0s

Each managed local model row gains an O(1) selector (auto / off / all / deepN),
exported to the child `cortiq serve` as CMF_O1. "off" forces exact attention,
curing incoherent output from a too-aggressive `--o1` conversion hint without
re-converting the file; "all"/"deepN" force-enable it. Auto (default) follows
the hint recorded in the .cmf. Surfaced in /admin/api/cmf per-server status.
This commit is contained in:
infosave2007
2026-07-14 12:32:51 +03:00
parent d93ac537c8
commit 851cc7f1cb
8 changed files with 29 additions and 6 deletions
Generated
+1 -1
View File
@@ -285,7 +285,7 @@ checksum = "1d07550c9036bf2ae0c684c4297d503f838287c83c53686d05370d0e139ae570"
[[package]]
name = "cortiq-gateway"
version = "0.2.18"
version = "0.2.19"
dependencies = [
"anyhow",
"arc-swap",