Text-to-Speech
English

Why is the StyleTTS2 fine-tuned checkpoint file size larger than the original?

#3
by umerahsan - opened

The original checkpoint file for StyleTTS2 is around 771 MB, but after fine-tuning, the checkpoint size increases significantly eg vokan itโ€™s around 2.04 GB. The model parameters and inference code remain unchanged, so what causes this increase in file size?

ShoukanLabs org

Hi, Those may have optimizer states for various modules inside this net. I've seen the original author of STTS sometimes removing those to make the weights smaller.

2.04GB is very natural.

Korakoe changed discussion status to closed

Hi, Yes the fine-tuned files do have optimizer states. I appreciate your reply. Thank you!

Sign up or log in to comment