Skip to content

RELEASING.md

Latest commit

 

History

History
57 lines (45 loc) · 1.5 KB

RELEASING.md

File metadata and controls

57 lines (45 loc) · 1.5 KB
Raw

Manual validation and release process

Validation

This repository does not run hosted CI or continuous deployment. Before a push or release, run the stable engine ABI build, core assembly tests, and the deterministic synthetic parity gates locally:

make -j"$(nproc)" engine-check
make -j"$(nproc)" test
make -j"$(nproc)" m35-check m38-check m58-check
make model-tool-check

RTF promotion must run on the fingerprinted target machine because hosted CPU topology, contention, and power policy are not controlled tightly enough for the performance policy in docs/PERFORMANCE.md.

Release

  1. Confirm the target-machine quality and performance gates.
  2. Update release notes and version-facing documentation.
  3. Build the archive and checksum locally.
  4. Create and push an annotated v* tag and publish the archive manually.
git tag -a v0.2.2 -m "DiffSinger-ASM v0.2.2"
git push origin v0.2.2

Create the archive locally with:

make package VERSION=v0.2.2

Release builders that need a portable binary must pass an explicit AVX2/FMA baseline instead of the local-development -march=native default. Optional AVX-VNNI kernels remain protected by runtime feature selection.

The archive contains:

bin/dsasm-acoustic
bin/dsasm-vocoder-m40
lib/libdsasm.so
include/dsasm_engine.h
config/best-inference.env
README.md
docs/ENGINE_ABI.md
docs/PERFORMANCE.md
docs/OPENUTAU_OFFLINE_MODEL_PROTOCOL.md
bin/dsasm-model-tool
share/dsasm-model-tool/app/...
share/dsasm-model-tool/python/...