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

about tango-full-ft-audiocaps #37

Open
Yusiissy opened this issue Oct 17, 2023 · 1 comment
Open

about tango-full-ft-audiocaps #37

Yusiissy opened this issue Oct 17, 2023 · 1 comment

Comments

@Yusiissy
Copy link

Hi, thanks for your great open source work!

In your work, I noticed that you used audiocaps dataset to fine tune on tango-full ckpt,
can I know the command for your fine tuning process?
Do I need to modify the learning rate (default=3e-5) and do I use --hf_model or --resume_from_checkpoint in the command?

Looking forward to your reply, thanks again!😊

@deepanwayx
Copy link
Collaborator

Yes, you can use the --hf_model argument to pass the tango-full model checkpoint for doing that. The full command would be:

accelerate launch train.py \
--train_file="data/train_audiocaps.json" --validation_file="data/valid_audiocaps.json" --test_file="data/test_audiocaps_subset.json" \
--hf_model "declare-lab/tango-full" --unet_model_config="configs/diffusion_model_config.json" --freeze_text_encoder \
--gradient_accumulation_steps 4 --per_device_train_batch_size=2 --per_device_eval_batch_size=2 --augment \
--learning_rate=3e-5 --num_train_epochs 40 --snr_gamma 5 \
--text_column captions --audio_column location --checkpointing_steps="best"

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

No branches or pull requests

2 participants