[Discussion] how to train the finetune dataset

Hey folks,

i've looked everywhere for some info on this but i cant seem to find anything..

the idea is to train the 2 stems finetune dataset (2stems-finetune).
i have some vocals and instrumentals and the mixdowns and have written most of them into a test_data.csv and some into a valid_data.csv.

i altered the musdb_config.json to match my .csv names.

now, where do i specify that i want to train an **existing** model and **not** make a new one?

when i use:
"model_dir": "musdb_model":
it starts to load all the audio and starts training.

i tried specifying "model_dir": "2stems-finetune":

INFO:spleeter:Start model training
INFO:spleeter:Model training done

It seems to be done in 2 seconds. No audio is loaded. yet there are 510 songs i am training on.

can anyone clarify what i am doing wrong or give any advice?

Start with musdb_config.json and the training configuration described in the issue, then trace how model_dir is interpreted when training the 2stems-finetune dataset. Check the existing training documentation or entry point for the distinction between creating and continuing a model; done means the workflow clearly explains the required configuration and successfully processes the supplied CSV datasets.

Written by the indexing model from the issue text.