Skip to content

v1.5.2

Choose a tag to compare

View all tags
@zmxnpquryet-gif zmxnpquryet-gif released this 23 Sep 11:22
· 31 commits to main since this release
v1.5.2
698b625

SDengine MoE optimization + default-llama.cpp routing.

  • Imported GGUF (incl. MoE) now defaults to llama.cpp; SDengine is strictly opt-in via Model Manager or the global runtime switch, with a one-time stale-prefs migration.
  • SDengine (TEST, MoE-only): fused streaming quantized matvec, RoPE cache, scratch reuse, heap topK, zero-copy expert paging, non-evicting prefetch, deferred-arch fail-fast.
  • Adversarial review rounds (codex/cmdc/agy), all findings fixed. Unit: engine 42/42, app 83/83 green.

versionCode 18 / versionName 1.5.2, signed with the v1.5.1 upload key (SHA-256 791f2c26...). Updates from v1.5.1 install cleanly; v1.4.0-or-earlier installs still need one reinstall (key rotation, see v1.5.1 notes).

Assets 3
Loading