Skip to content
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.

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

More scripts #61

Merged
merged 7 commits into from
May 14, 2020
Merged

More scripts #61

merged 7 commits into from
May 14, 2020

Conversation

prafullasd
Copy link
Collaborator

No description provided.

@prafullasd prafullasd requested a review from heewooj May 10, 2020 22:21
models the lyrics, and we use its last layer to produce keys/values that are attened to by the decoder transformer
- Single Encoder Decoder: This is a simplification where we combine them into a single model. We merge the text vocab
and VQ vocab into a single large vocab, and the lyric tokens and VQ tokens into a single longer sequence of tokens which
we autoregressively model together.
Copy link
Contributor

Choose a reason for hiding this comment

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

👍

README.md Outdated
mpiexec -n {ngpus} python jukebox/train.py --hps=vqvae,small_prior,all_fp16,cpu_ema --name=pretrained_vqvae_small_prior_labels \
--sample_length=1048576 --bs=4 --aug_shift --aug_blend --audio_files_dir={audio_files_dir} \
--labels=True --train --test --prior --levels=3 --level=2 --weight_decay=0.01 --save_iters=1000 \
--labels_v3=True --y_bins=({artists},{genres}) --max_bow_genre_size=1 --min_duration=60.0 --max_duration=600.0 --t_bins=64
Copy link
Contributor

Choose a reason for hiding this comment

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

might be useful to add that --min_duration --max_duration should be decided based on the dataset people have, and that --min_duration should be at least ~23.8 seconds so that there's always a context we can train on.

n_ctx=6144,
prior_width=1024,
prior_depth=48,
heads=4,
Copy link
Contributor

Choose a reason for hiding this comment

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

nitpick: in general, lgtm, but perhaps, 2 heads will be enough? 1B lyrics uses 2 heads as well

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

good point, changing!

Copy link
Contributor

@heewooj heewooj left a comment

Choose a reason for hiding this comment

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

LGTM, left some comments

@prafullasd prafullasd merged commit 085eb81 into master May 14, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants