Can't "Train Model" in Mainline Kaggle.

[manic slate]

All the processes before pressing the button of "Train Model" were fine in Hina's RVC Mainline Kaggle, but it gave me errors especially with the bottom text in the train.log

--
"magic_number = pickle_module.load(f, **pickle_load_args)
_pickle.UnpicklingError: invalid load key, '<'."

[uncut fern]

How did you upload the dataset? Through the kaggle ui?

[manic slate]

I uploaded the dataset of 2 minutes and 30 seconds, with ImJoy Finder URL provided.
And the “dataset” folder was locked, so I added other folder and uploaded the .wav file there.

[uncut fern]

Ye not quite sure too imma take a look at it later when I'm home.. prob 7 hours from now tho

If you really need a model ig use applio for the time being

[manic slate]

Understood.
I'll wait while going for applio.

[uncut fern]

tested it and it seems to be working fine

the only thing i can think of is the pretrain

how did you add/download the model?

[manic slate]

By following the tutorial of using RVC Mainline Kaggle in Rentry, I inputted two codes of “!wget https://huggingface.co/blaise-tk/TITAN/resolve/main/models/medium/48k/pretrained/D-f048k-TITAN-Medium.pth” and “!wget https://huggingface.co/blaise-tk/TITAN/resolve/main/models/medium/48k/pretrained/G-f048k-TITAN-Medium.pth” between second cell and third cell.

[uncut fern]

hm ye not sure then

tried mine with the same thing and works no problem

[manic slate]

Hmm, I’ll try and put two pre-trained models of TITAN on created folder if it works.

And somehow, the original “dataset” folder on the file url is locked.

Might be ngrok problem I guess?

[odd dove]

I'd personally recommend the default pretrain or KLM 4.3 for singing

the recent applio has quite lot of improvements for training and some new refinegan and mrf pretrains only work on the latest applio

[manic slate]

Saw pretrains like KLM in this server and huggingface too.
I’ll look forward to them in the future.

[uncut fern]

its locked cause its a "linked" folder where that dataset folder is a fake folder that doesnt exist, it gets filled by the input dataset part from kaggle

this