training/finetune information

#1
by tp899 - opened

Would you share how this model was trained/fine-tuned?
Training script, training data, batch-size and number of iterations

The model was trained using the scripts in this repo:
https://github.com/stlohrey/chatterbox-finetuning/tree/master/src
training data was https://huggingface.co/datasets/MrDragonFox/DE_Emilia_Yodas_680h .
I just finetuned for 1 epoch, batchsize 4 (*2 gradient accumulation) for the t3 part and batchsize 8 for s3gen/flow.

How long does it take for you to complete 1 epoch? I'm trying to run it using one A100 GPU but it takes more than 150 hours to complete just one epoch.

Depends on your data and settings, I would need more info on this issue...

I solved the issue. Thanks

Sign up or log in to comment