search

rakuri255 / UltraSinger

AI based tool to convert vocals lyrics and pitch from music to autogenerate Ultrastar Deluxe, Midi and notes. It automatic tapping, adding text, pitch vocals and creates karaoke files.
MIT License
290 stars 26 forks source link

Docker Support with GPU #148

Closed binarygeek119 closed 5 months ago

binarygeek119 commented 5 months ago

i have created a dockerfile that does build a image that will work with a GPU pass thought to run this docker you need to run this command docker build -t ultrasinger .

then after the build run.

docker run --gpus all -it --name UltraSinger ultrasinger

right now this drops you into the docker commandline. all you have to do is run python3 UltraSinger.py -i file.

right now i edited the dockerfile to just copy a mp3 for testing and all works fine for me. rightnow to use this you must run this from commandline. but its all self contained. to get this in and out run this command instead. docker run --gpus all -it --name UltraSinger -v $(pwd)/mp3:/mp3 ultrasinger this will let you get files in and out of the docker.

the next step would be more testing and then making a webui to pass eveything though.

im sure there more needed then what i came up with, this is my first docker ive build.

thoughts?

binarygeek119 commented 5 months ago

While this does produce a file that is useable, it doesn't use the GPU for everything. Some dependency not ment In the docker. Still playing with it. I may have a idea what's wrong but it's late and I need to go to bed. Will keep working on it.

rakuri255 commented 5 months ago

While this does produce a file that is useable, it doesn't use the GPU for everything. Some dependency not ment In the docker. Still playing with it. I may have a idea what's wrong but it's late and I need to go to bed. Will keep working on it.

Cool! will try it. You can try upgrade pytorch and Tensorflow. The Tensorflow updgrade was needed for Colab, since they used Cuda 12.1. Maybe its a similar issue?

!pip3 install torch==2.3.1+cu121 torchvision==0.18.1+cu121 torchaudio==2.3.1+cu121 --index-url https://download.pytorch.org/whl/cu121
!pip install tensorflow[and-cuda]==2.15.0
binarygeek119 commented 5 months ago

i have all steps working on gpu now never seen it go so fast. it does take a bit to build. do you want me to upload to docker for people to pull from? i do know there are ways you can link github to rebuild the docker when i update come out. for this i have no idea how to do it. but i would be happy to upload a new docker image when updates come out.

binarygeek119 commented 5 months ago

i have tested it about 5 times now and it seems to be put together right. i need other to test to see any issues.

rakuri255 commented 5 months ago

i have all steps working on gpu now never seen it go so fast. it does take a bit to build. do you want me to upload to docker for people to pull from? i do know there are ways you can link github to rebuild the docker when i update come out. for this i have no idea how to do it. but i would be happy to upload a new docker image when updates come out.

Nice! Good work! We need an github CI for this. Its not that hard -> publishing-docker-images We can discusse in discord about that and try it out.

Can you describe the docker changes a bit in the readme? Then i accept the PR.

rakuri255 commented 5 months ago

Check CodeFactor — 7 issues found.

binarygeek119 commented 5 months ago

testing the updated dockerfile with the fixes

binarygeek119 commented 5 months ago

i think i got it now just have to build

binarygeek119 commented 5 months ago

it fully works and no errors.