Skip to content

Preload cuDNN/cuBLAS for the WhisperX encoder #14

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
hollygrimm merged 1 commit into from
Jul 17, 2026
Merged

Preload cuDNN/cuBLAS for the WhisperX encoder #14

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
14 changes: 14 additions & 0 deletions scripts/transcribe_worklist.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -141,6 +141,20 @@ def main() -> int:
wav_dir = args.work_dir / "wav"
wav_dir.mkdir(exist_ok=True)

# ctranslate2 dlopens cuDNN by soname; pip-installed nvidia wheels are not
# on the loader path, so preload them (RTLD_GLOBAL) or the whisper encoder
# aborts with "Unable to load libcudnn_cnn.so.9".
import ctypes
import site

for sp in site.getsitepackages():
for pattern in ("nvidia/cublas/lib/libcublas*.so*", "nvidia/cudnn/lib/libcudnn*.so.9"):
for lib in sorted(Path(sp).glob(pattern)):
try:
ctypes.CDLL(str(lib), mode=ctypes.RTLD_GLOBAL)
except OSError:
pass

# Heavy import deferred so dry runs work without GPU deps loaded.
from journal_utilities.transcribe.transcribe import TranscriptionService

Expand Down