Model Size of fine tuned Model

[deguodedongxi]

Hello everyone,

I was finetuning a multilingual Model. I used the base model from hugging face, which is ~0.77Gb in size. I finetuned first with one new speaker and than with two speaker. Every checkpoint and also the final model with just one extra speaker is ~1.9Gb in size. With 2 speakers it is ~2.1Gb in size.

Inside the code I could not really make out a reason, why the model is gaining in size the more speakers I add.
Could you help me understand the reason, why it is getting bigger and bigger? I couldnt find a proper reason for that inside the code.

The base model was trained on the LibriTTS Dataset, which is pretty big, that's why the jump from 0.700gb to 2.1gb with a finetuned model doesnt really make sense to me.

Thanks in advance!