High RAM usage acoustic model

After MFA training, I've got 39778 Textgrids and 39778 wav file. The issue that when I ran acoustic trainer it cost 14GB RAM and keep increasing. How to fix this? Thank you

I've tried to decrease batch size to 16 but it's not work

There are two ways to resolve this issue:

use machines with more RAM. Google Colab allows users to select machine with more RAM at Runtime -> Change runtime type
modify the following function to prevent it loading all data to memory.

vietTTS/vietTTS/nat/data_loader.py

Line 85 in 346d467

    
           def load_textgrid_wav(data_dir: Path, token_seq_len: int, batch_size, pad_wav_len, mode: str):

How about this solution: shuffle data and divide into 2-3 part and training sequentially each part ?
Does this solution lead to worse result or not ?