Labyricorn/voicebox
1
0
Fork 0
mirror of https://github.com/jamiepine/voicebox.git synced 2026-09-26 13:45:16 -07:00
Code Issues Packages Projects Releases Wiki Activity

fix(mlx): bundle native libs and broaden error handling for Apple Silicon

Browse Source 
The distributed macOS aarch64 binary shipped without MLX acceleration despite
the model and backend code supporting it. Two root causes:

1. **OSError not caught in platform_detect.py**
   PyInstaller bundles isolate the filesystem, so when MLX tries to load its
   Metal shader libraries (.metallib) it raises OSError, not ImportError.
   platform_detect.get_backend_type() only caught ImportError, causing a
   silent fallback to PyTorch even on Apple Silicon hardware.
   Fix: broaden the except clause to (ImportError, OSError, RuntimeError)
   and import mlx.core instead of mlx (forces native lib loading eagerly).

2. **collect_data_files used instead of collect_all for MLX**
   build_binary.py and voicebox-server.spec used --collect-data /
   collect_data_files for mlx and mlx_audio. This copies Python source and
   pure-Python data, but NOT native shared libraries (.dylib, .metallib).
   Fix: switch to --collect-all / collect_all which captures binaries too,
   then pass them to Analysis(binaries=...) in the spec.

Result: macOS Apple Silicon users now get MLX inference (~4-5x faster than
PyTorch CPU), matching the performance documented in the README.
This commit is contained in:
Eva
2026-02-18 16:51:48 +01:00
parent eb2cd861b1
commit 829d4d6d5b
3 changed files with 22 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/voicebox-server.spec
+8 -3
View File
@@ -6,8 +6,13 @@ from PyInstaller.utils.hooks import copy_metadata
datas = []
hiddenimports = ['backend', 'backend.main', 'backend.config', 'backend.database', 'backend.models', 'backend.profiles', 'backend.history', 'backend.tts', 'backend.transcribe', 'backend.platform_detect', 'backend.backends', 'backend.backends.pytorch_backend', 'backend.utils.audio', 'backend.utils.cache', 'backend.utils.progress', 'backend.utils.hf_progress', 'backend.utils.validation', 'torch', 'transformers', 'fastapi', 'uvicorn', 'sqlalchemy', 'librosa', 'soundfile', 'qwen_tts', 'qwen_tts.inference', 'qwen_tts.inference.qwen3_tts_model', 'qwen_tts.inference.qwen3_tts_tokenizer', 'qwen_tts.core', 'qwen_tts.cli', 'pkg_resources.extern', 'backend.backends.mlx_backend', 'mlx', 'mlx.core', 'mlx.nn', 'mlx_audio', 'mlx_audio.tts', 'mlx_audio.stt']
datas += collect_data_files('qwen_tts')
datas += collect_data_files('mlx')
datas += collect_data_files('mlx_audio')
# Use collect_all (not collect_data_files) so native .dylib and .metallib
# files are bundled as binaries, not data. Without this, MLX raises OSError
# when loading Metal shaders inside the PyInstaller bundle.
from PyInstaller.utils.hooks import collect_all as _collect_all
_mlx_datas, _mlx_bins, _mlx_hidden = _collect_all('mlx')
_mlxa_datas, _mlxa_bins, _mlxa_hidden = _collect_all('mlx_audio')
datas += _mlx_datas + _mlxa_datas
datas += copy_metadata('qwen-tts')
hiddenimports += collect_submodules('qwen_tts')
hiddenimports += collect_submodules('jaraco')
@@ -18,7 +23,7 @@ hiddenimports += collect_submodules('mlx_audio')
a = Analysis(
['server.py'],
pathex=[],
binaries=[],
binaries=_mlx_bins + _mlxa_bins,
datas=datas,
hiddenimports=hiddenimports,
hookspath=[],