PyTorch out of CPU memory?

[jeffcrouse]

When running the first sampling command:

python jukebox/sample.py --model=5b_lyrics --name=sample_5b --levels=3 --sample_length_in_seconds=20 \
--total_sample_length_in_seconds=180 --sr=44100 --n_samples=6 --hop_fraction=0.5,0.5,0.125

I get this error

RuntimeError: [enforce fail at CPUAllocator.cpp:56] posix_memalign(&data, gAlignment, nbytes) == 0. 12 vs 0

According to this issue, that means I am out of CPU memory. But I am using a P5000 Paperspace machine with 30GB of CPU memory, which I assumed would be more than enough. How much is required?

If this isn't sufficient, can anyone suggest a cloud provider that has a machine powerful enough to run jukebox?

Thanks!

[youdontown]

I would like some system requirements because I am running out of RAM as well (I have 24 GB).

It looks like people are recommending Google Colab as a cloud service provider though. There's a link in the README.

[jeffcrouse]

Yes, FWIW, I signed up for Google Collab Pro and I'm having more luck than on a Paperspace server. Slower, but about 100x less expensive, so that's fine with me for now.

I originally found the Collab link in this issue #40 This version apparently "adds support for primed audio by supplying an audio file in your Google Drive"