Skip to content

StaleyNET v0.1.1

Choose a tag to compare

@arpanpathak arpanpathak released this 04 Nov 07:08
· 95 commits to main since this release
e2aa933

StaleyNET v0.1.1

Initial tagged release of the rebuilt StaleyNET repository.

Highlights

  • Clean 33-PR history, rebuilt module by module.
  • 219 files covering code, paper, book, studio, configs, and artifact metadata.
  • 171 tests pass, including 18 GPU integration tests on a Jetson Orin.
  • CI covers rustfmt, clippy with warnings denied, tests, cargo-deny, and the 99 percent audio coverage gate.
  • staleynet-audio 0.1.1 is published on crates.io with a crate-specific README.
  • The private Hugging Face model and dataset cards were rewritten for this release.

Install the audio crate

cargo add staleynet-audio

Repository layout

  • crates/staleynet-audio: CPU decoding, resampling, log-mel, Griffin-Lim, WAV.
  • crates/staleynet-cuda: cuda-oxide kernels and shape-checked operations.
  • crates/staleynet-model: Transformer, optimizer, decoding, checkpoints.
  • crates/staleynet-data: catalogues, cache store, RVQ tokenizer, sampler.
  • crates/staleynet-train: training loop and checkpoint policy.
  • crates/staleynet-cli: the staleynet command.
  • paper/: LaTeX sources and compiled PDFs.
  • book/: mdBook source and figures.
  • studio/: browser studio for generation and recording.
  • artifacts/: run metadata, checksums, and Hugging Face manifest.

Crates.io status

staleynet-audio is published. The GPU crates remain source-only for now because they depend on cuda-oxide through git revisions, and crates.io rejects git dependencies. They can be packaged once cuda-oxide publishes the host and device crates, or on a dedicated packaging branch.

Verification

  • cargo fmt --all -- --check
  • cargo clippy --workspace --all-targets -- -D warnings
  • cargo deny check
  • cargo test -p staleynet-audio
  • cargo oxide test -- --release --workspace on the Jetson Orin
  • Audio line coverage: 99.10 percent

Links

License

Code is Apache-2.0. Paper and book are CC BY 4.0. Checkpoints and derived data are CC BY-SA 4.0 subject to the upstream dataset terms.