Skip to content

Update changelog: fix WhisperX installation, dependency pins, and GPU… #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chris-colinsky merged 1 commit into from
Mar 2, 2026
+8 −1
Merged

Update changelog: fix WhisperX installation, dependency pins, and GPU… #1

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 8 additions & 1 deletion CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

## [Unreleased]

## [0.1.0] - 2026-02-27
## [0.1.0] - 2026-03-01

### Added

Expand All @@ -28,5 +28,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
- Scratch space management with automatic cleanup of intermediate files
- GPU performance-based ordering via `gpu_tflops.toml` lookup table

### Fixed

- WhisperX model loading: pinned install to commit `741ab9a` (v3.7.6) — the `v3.1.1` tag uses an older API (`transcribe.py`) that lacks the `device_index` parameter required by the ctranslate2 backend
- `make install-whisperx` now also installs WhisperX runtime deps (`av`, `ctranslate2`, `faster-whisper`, `flatbuffers`, `nltk`, `onnxruntime`) which were previously missing after a `--no-deps` install
- `transformers` capped at `<4.40.0` — versions 4.40+ use `torch.utils._pytree.register_pytree_node`, an API introduced in PyTorch 2.2, which breaks with the pinned PyTorch 2.1.2
- `make dev-setup` now reinstalls CUDA torch wheels (`torch==2.1.2+cu121`, `torchaudio==2.1.2+cu121`) as its final step — `uv sync` resolves torch from PyPI and installs the CPU-only build, silently breaking GPU inference

[Unreleased]: https://github.com/LunarCommand/audio-refinery/compare/v0.1.0...HEAD
[0.1.0]: https://github.com/LunarCommand/audio-refinery/releases/tag/v0.1.0