3.1.0 — pure pip install: madmom replaced by madmom-infer

• madmom replaced by madmom-infer: installation is now plain pip install all-in-one-infer — no git installs, no post-install hooks. Verified: correctly-rounded STFT magnitude (1 ULP fix over madmom's np.abs bug) closure-proven, and end-to-end on 3 real songs all discrete outputs (bpm/beats/downbeats/segments) were exactly identical
• Fixed a fresh-install crash on torchaudio ≥2.11 (missing torchcodec decoders): wav/flac input loading now goes through soundfile (proven bit-identical), mp3 stays on torchaudio
• demucs-infer bumped to >=4.2.2 (brings its own torchaudio-2.11 fix)
• Revived 3 E2E tests (test_analyze/test_sonify/test_visualize) that were silently dead since the 3.0.0 rename (stale imports)
• Removed setup.py — its post-install madmom hook never actually fired under pip/uv

v3.0.1 — activation_fps field, README rename announcement

Added

• AnalysisResult.activation_fps — the frame rate (Hz) of the activations array, populated whenever activations are requested, so downstream consumers can map activation frame indices to timestamps (#2). Backward compatible: older saved JSONs load with None.

Documentation

• Official rename announcement (all-in-one-fix → all-in-one-infer) now front and center in the README; stale allin1 imports in current-usage examples fixed; obsolete NATTEN-required claims replaced with the pure-PyTorch reality; opt-in performance flags documented in CLI and Python API sections.

v3.0.0 — all-in-one-infer: NATTEN-free, 25% faster, one-click install

First release under the new name all-in-one-infer (formerly all-in-one-fix; Python package renamed allin1fix → allin1_infer — breaking import change, hence the major version).

Highlights

• NATTEN dependency removed: neighborhood attention reimplemented in pure PyTorch, verified bit-identical to NATTEN 0.17.5 (golden fixtures, layer-level A/B, and end-to-end analyze() comparison on real music — beats/downbeats/segments all exactly equal). Installs with a single pip install on any torch ≥ 2.0 (upper bound lifted), CPU-only machines, macOS/Apple Silicon, and Windows. No C++ toolchain. NATTEN stays available as an optional [natten] extra for fused GPU kernels.
• ~25% faster inference, bit-identical output: TF32 matmul, parallel ensemble checkpoint downloads, in-memory stem pipeline.
• Inference-only: training code removed (see upstream mir-aidj/all-in-one for training).
• Opt-in performance knobs (compile_model, demucs_overlap, demucs_fp16) — default off, accuracy-affecting ones clearly documented.
• Deep-module refactor + navigation headers throughout; publish workflow gates on tests.

Install

pip install all-in-one-infer
pip install git+https://github.com/CPJKU/madmom   # last remaining extra step (madmom-infer, its pure-Python replacement, is in development)

v2.0.4 - PyTorch Compatibility Fix

Fix: PyTorch Compatibility Constraint

This release fixes compatibility issues with PyTorch 2.8+ by constraining the PyTorch version requirement.

Changes

• Fixed: Constrained PyTorch to >=2.0.0,<2.8.0 to prevent compatibility issues with natten 0.17.5
• Reason: PyTorch 2.8+ removed internal _device_t API that natten 0.17.5 depends on
• Updated: README with compatibility warnings and version constraints

Compatibility

• ✅ PyTorch 2.0-2.7: Fully supported
• ❌ PyTorch 2.8+: Not compatible (requires natten 0.21+ support - coming in future release)

Installation

pip install torch>=2.0.0,<2.8.0
pip install all-in-one-fix --no-build-isolation

v2.0.3 - Fix Missing Subpackages

Bug Fixes

• Fixed missing subpackages: Changed package discovery to include all subpackages (models/, postprocessing/, training/) in PyPI distribution
• This fixes the ModuleNotFoundError: No module named 'allin1fix.models' error when installing from PyPI

Changes

• Updated pyproject.toml to use automatic package discovery with {find = {}}
• Ensures all subpackages are properly included in the wheel distribution

Installation

pip install torch>=2.0.0
pip install all-in-one-fix --no-build-isolation
pip install git+https://github.com/CPJKU/madmom  # If auto-install fails

v2.0.2 - PyPI Project Name Consistency

What's New

• ✨ Consistent Naming: PyPI project name now matches repository name ()
• 🔧 Python Package: Still uses for imports (no code changes needed)
• 📚 Documentation: Updated all references to reflect new PyPI project name

Installation

pip install torch>=2.0.0
pip install all-in-one-fix --no-build-isolation

Note:

• Install with: pip install all-in-one-fix (PyPI project name)
• Import with: import allin1fix (Python package name)
• CLI: allin1fix --help (based on Python package name)

Changes

• Changed PyPI project name from allin1fix to all-in-one-fix (matches repository)
• Updated all documentation and installation instructions
• Python package name remains allin1fix (backward compatible for imports)
• Updated PyPI URLs and trusted publishing configuration

Consistency

Now matches the pattern used by other projects:

• ✅ PyPI project name: all-in-one-fix (matches repository)
• ✅ Repository name: all-in-one-fix
• ✅ Python package: allin1fix (for imports)

v2.0.1 - Auto-install madmom

What's New

• ✨ Auto-install madmom: madmom is now automatically installed from git during package installation
• 🔧 Build system: Switched to setuptools backend to support post-install hooks
• 📚 Documentation: Updated installation instructions to reflect automatic madmom installation
• 🐛 Fix: Version now correctly reads from about.py

Installation

pip install torch>=2.0.0
pip install allin1fix --no-build-isolation

madmom will be automatically installed during the allin1fix installation process.

Changes

• Added post-install hook in setup.py to auto-install madmom from git
• Updated README.md with simplified installation instructions
• Fixed version reading for setuptools build backend
• Improved error messages if madmom installation fails

v2.0.0 - PyPI Ready Release

🎉 First PyPI Release

This release makes allin1fix available on PyPI with comprehensive installation documentation.

✨ Key Features

• 📦 Available on PyPI: pip install torch>=2.0.0 && pip install allin1fix --no-build-isolation
• 🔧 All dependencies install from PyPI (demucs-infer)
• 📚 Comprehensive installation documentation
• 🚀 Ready for production use

📝 Installation

pip install torch>=2.0.0 && pip install allin1fix --no-build-isolation

🔧 Changes

• Updated README with PyPI installation instructions
• Removed worzpro references from documentation
• Updated pyproject.toml with proper maintainer information
• Added PyPI publishing documentation
• Updated GitHub workflow for automated PyPI publishing
• Consolidated all documentation into single README.md
• Fixed build configuration (removed torch from build-system.requires)

📦 Package Information

• Version: 2.0.0
• Python: 3.9+
• PyTorch: 2.0.0+
• License: MIT