breadlicker45

/

musenet-untrained

like 0

Hi,

I want to train the model on a collection of midi files. Some hints on how to do it? Sadly I couldn't find anything about it on the web.

i wrote code that mass converts it on a website that converts midi to musenet enocders and stores the in database. so like this, midi files -> use website to convert to musenet enocders and store in db -> json -> cleaned json -> csv -> cleaned csv -> upload dataset

here is the code to convert it on the site.
https://pastebin.com/wz5T5Wa3
and you will need to download this and host it locally https://github.com/MrCheeze/musenet-midi
and then you will need this to make a list of the midi files in your dataset https://pastebin.com/tj5xQ78f

make a list of your midi files and put it in a txt https://pastebin.com/tj5xQ78f -> then host the site https://github.com/MrCheeze/musenet-midi -> then run this code on the hosted site in the dev consle
https://pastebin.com/wz5T5Wa3 -> download the db in a json file from your browser -> then clean any blank rows from the json file aka the db -> then convert it to csv -> then clean the csv -> upload csv -> train model like normal llm using transfomers or whatever you use to train models with

Thank you for the detailed explanation, it is a very good starting point! This is gonna be my project for this week. :)

Thank you for the detailed explanation, it is a very good starting point! This is gonna be my project for this week. :)

np, you are going to need a lot of ram for running the code in the dev console
around 32GB is good i think

i do highly suggest you train this "https://huggingface.co/breadlicker45/neox-musenet-untrained" and not "https://huggingface.co/breadlicker45/musenet-untrained"