Wolf O'Donnel (Japanese) / Mangio-Crepe / 43 Epochs | AI HUB | Page 1

hollow raft Oct 16, 2024, 9:37 PM

#

Download: https://huggingface.co/lyery/mymodels/resolve/main/wolf_43e_2494s.rar?download=true

arf woof rarw cutedog
too lazy to find japanese samples

Info:
Dataset: 3:37 minutes (no, this does not use ssbu samples despite this post pic)
Batch size: 2
Optimizer: Prodigy
Prodigy settings: lr=1,
betas=(0.9, 0.99),
decouple=True,
weight_decay=0.01,
d_coef=0.8,
use_bias_correction=True,
safeguard_warmup=True
Pitch extractor: Mangio-crepe
Hop length: 32
Sample rate: 32k
Steps: 2494
Voice actor: Kōsuke Takaguchi

half socketBOT Oct 16, 2024, 9:37 PM

#

Rate the model

To rate this model, either use /rate or click the buttons below.

half socketBOT Oct 16, 2024, 9:37 PM

#

half socket

uncut sigil Oct 16, 2024, 10:10 PM

#

wat

teal ore Oct 16, 2024, 10:19 PM

#

uncut sigil wat

He changed the optimizer rvc uses from AdamW to Prodigy

uncut sigil Oct 16, 2024, 10:20 PM

#

teal ore He changed the optimizer rvc uses from AdamW to Prodigy

how? shocked

teal ore Oct 16, 2024, 10:22 PM

#

#✨│ai-help message

uncut sigil Oct 16, 2024, 10:22 PM

#

is prodigy really worth it?

#

i've seen that noobies said that takes longer to train or idk

teal ore Oct 16, 2024, 10:22 PM

#

thats with Lion

#

idk about others

uncut sigil Oct 16, 2024, 10:22 PM

#

maybe i'll try it ig

#

to see if it has any advantage on training

teal ore Oct 16, 2024, 10:23 PM

#

well im currently messing with Adabelief and the training times are about the same

#

maybe like 10-20 seconds slower

uncut sigil Oct 16, 2024, 10:25 PM

#

yea i'll just use the default optimizer

#Wolf O'Donnel (Japanese) / Mangio-Crepe / 43 Epochs