Skip to content

Fix MPS tensor dtype conversion in audio decoding #20

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
lmangani merged 2 commits into from
Jan 30, 2026
Merged

Fix MPS tensor dtype conversion in audio decoding #20

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
75fb81b
Initial plan
Copilot Jan 30, 2026
10b42e4
Fix MPS tensor dtype issue: explicitly preserve torch.long for frame …
Copilot Jan 30, 2026
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
3 changes: 2 additions & 1 deletion backend/app/services/music_service.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -846,7 +846,8 @@ def _pad_audio_token(token):

# Move frames to codec device (keep dtype as long for indexing)
# frames contains token IDs (integers) used as indices, so dtype must remain long
frames_for_codec = frames.to(device=pipeline.codec_device)
# Explicitly preserve torch.long dtype when moving to device (critical for MPS)
frames_for_codec = frames.to(device=pipeline.codec_device, dtype=torch.long)
wav = pipeline.codec.detokenize(frames_for_codec)

# Cleanup codec if using lazy loading (free VRAM for next generation)
Expand Down