Skip to content

Fix mel spectrogram preprocessor allocating gigabytes of planned memory #18229

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
mergennachin merged 1 commit into from
Mar 17, 2026
Merged

Fix mel spectrogram preprocessor allocating gigabytes of planned memory #18229

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
11 changes: 8 additions & 3 deletions extension/audio/mel_spectrogram.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -192,10 +192,15 @@ def export_processor(model=None, output_file="whisper_preprocess.pte"):
if model is None:
model = WhisperAudioProcessor()

audio_tensor = torch.randn(93680)
if model.streaming:
# Streaming processes small windows per step. 2 seconds gives
# comfortable headroom while keeping the memory plan tight.
max_samples = 2 * model.sampling_rate
Copy link
Contributor

@digantdesai digantdesai Mar 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Any performance issues with this? In the streaming mode, each inference takes 2s worth of samples and start over again for next two seconds?

Copy link
Contributor Author

@mergennachin mergennachin Mar 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No performance issue. The max_samples = 2 * sampling_rate is only the dynamic shape upper bound at export time. It tells the memory planner the maximum buffer size to allocate. It doesn't affect how inference runs.

At runtime, the streaming preprocessor is called with ~1,640 samples per step (~0.1s). The exported graph handles any input size from 1 up to the declared max.

The 2-second cap just means if someone somehow passed more than 32,000 samples in a single call, it would fail. In practice the streaming window is fixed at 1,640 samples.

else:
max_samples = model.max_audio_len * model.sampling_rate
audio_tensor = torch.randn(min(93680, max_samples))
shapes_collection = torch.export.ShapesCollection()
max_n_chunks = int(model.max_audio_len * model.n_samples)
shapes_collection[audio_tensor] = {0: Dim.DYNAMIC(max=max_n_chunks)}
shapes_collection[audio_tensor] = {0: Dim.DYNAMIC(max=max_samples)}
with torch.no_grad(), torch.fx.experimental._config.patch(
backed_size_oblivious=True
):
Expand Down
Loading