Release 1.1.0

.github/workflows/release.yml

@@ -0,0 +1,89 @@
+name: Release
+
+# Push a tag like v1.1.0 to build the package, create the GitHub release (notes
+# taken from that version's CHANGELOG.md section) and publish to PyPI.
+on:
+  push:
+    tags: ["v*"]
+
+permissions:
+  contents: read
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Check the tag matches the package version
+        run: |
+          version=$(python -c "import tomllib; print(tomllib.load(open('pyproject.toml', 'rb'))['project']['version'])")
+          if [ "v$version" != "$GITHUB_REF_NAME" ]; then
+            echo "::error::Tag $GITHUB_REF_NAME does not match pyproject.toml version $version"
+            exit 1
+          fi
+
+      - name: Build and check the distributions
+        run: |
+          python -m pip install build twine
+          python -m build
+          twine check --strict dist/*
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  github-release:
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - name: Extract this version's notes from CHANGELOG.md
+        run: |
+          version="${GITHUB_REF_NAME#v}"
+          awk -v heading="## [$version]" '
+            /^## \[/ { if (found) exit; if (index($0, heading) == 1) { found = 1; next } }
+            found { print }
+          ' CHANGELOG.md > notes.md
+          if [ ! -s notes.md ]; then
+            echo "::error::CHANGELOG.md has no section for $version"
+            exit 1
+          fi
+
+      - name: Create the GitHub release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          gh release create "$GITHUB_REF_NAME" dist/* \
+            --title "$GITHUB_REF_NAME" --notes-file notes.md --verify-tag
+
+  pypi:
+    needs: build
+    runs-on: ubuntu-latest
+    # PyPI trusted publishing: the project must list this repository,
+    # workflow (release.yml) and environment (pypi) as a trusted publisher.
+    environment:
+      name: pypi
+      url: https://pypi.org/project/vayu-whisper/
+    permissions:
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - uses: pypa/gh-action-pypi-publish@release/v1

CHANGELOG.md

@@ -0,0 +1,63 @@
+# Changelog
+
+All notable changes to this project are documented here. The release workflow
+publishes the section for each version as its GitHub release notes.
+
+## [1.1.0] - 2026-10-01
+
+### Fixed
+
+- Batched decoding (`batch_size > 1`) no longer drops text at the end of each
+  30-second window, and every segment records its own window's `seek` (#3).
+- `word_timestamps=True` now works with batched decoding, the default for
+  `LightningWhisperMLX` (#4).
+- Batched temperature fallback follows the `temperature` schedule and
+  re-checks each step; a single temperature means no fallback (#5).
+- The rule that stops timestamps going backwards during decoding now applies
+  (#6).
+- CLI: each input file gets its own output name instead of every file
+  overwriting the first one's output (#7).
+- CLI: missing files, a missing ffmpeg and undecodable audio are reported
+  clearly and listed in the failure summary (#7).
+- Speculative decoding runs, uses a key/value cache and per-model
+  spectrograms, and gives exactly the target model's greedy output (#8).
+
+### Added
+
+- `AudioLoadError`, raised when ffmpeg is missing or cannot decode the audio.
+- `scripts/benchmark.py` to measure the batched speed-up on your own Mac (#11).
+- README sections on loading local models and on batched vs sequential
+  decoding (#11).
+- CI that runs the test suite on Linux and Apple Silicon (#2).
+
+### Changed
+
+These may need a change in code that calls Vayu:
+
+- `load_audio()` raises `FileNotFoundError` for a missing file (previously
+  `ValueError`), and `transcribe()` checks the path before loading the model.
+- `quant` raises `ValueError` when it can't be applied (an unknown level, a
+  model without a quantised build, or a repo path) instead of silently loading
+  full precision (#9).
+- `beam_size` and `patience` raise `NotImplementedError` up front; beam search
+  is not implemented. The CLI's `--patience` option is removed (#10).
+- `hallucination_silence_threshold` with `batch_size > 1` now warns that it is
+  ignored.
+- Speculative decoding's default pair is distil-large-v3 → large-v3; draft and
+  target models must share a vocabulary (#8).
+- Decoded timestamps can differ slightly from 1.0.0 because the timestamp rule
+  now applies (#6).
+- Development dependencies live in the `dev` extra only: `pip install -e
+  ".[dev]"` or `uv sync --extra dev` (#2).
+
+### Deprecated
+
+- `parallel_chunk_transcribe`: it never ran chunks in parallel. Use
+  `transcribe(audio, batch_size=N)` (#8).
+
+## [1.0.0] - 2026-01-19
+
+First release on PyPI as `vayu-whisper`.
+
+[1.1.0]: https://github.com/CodeWithBehnam/vayu/compare/v1.0.0...v1.1.0
+[1.0.0]: https://github.com/CodeWithBehnam/vayu/releases/tag/v1.0.0

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "vayu-whisper"
-version = "1.0.0"
+version = "1.1.0"
 description = "The fastest Whisper speech-to-text on Apple Silicon. 3-5x faster transcription via MLX batched decoding on M1/M2/M3/M4 Macs."
 readme = "README.md"
 license = {text = "MIT"}

tests/test_version.py

@@ -0,0 +1,9 @@
+"""The package's __version__ must match the version in pyproject.toml."""
+
+from importlib.metadata import version
+
+import whisper_mlx
+
+
+def test_version_matches_installed_metadata():
+    assert whisper_mlx.__version__ == version("vayu-whisper")

whisper_mlx/__init__.py

@@ -74,7 +74,7 @@
     parallel_chunk_transcribe,
 )
 
-__version__ = "1.0.0"
+__version__ = "1.1.0"
 
 __all__ = [
     # Main transcription function