setup notes - setup script is tbd

environment setup

install / update conda
download / clone this repository
create environment / activate environment
- conda env create -f environment.yml
  - alternatively use conda.lock.yml for exact package versions
- conda activate dualdiffusion
copy .env.default to .env, edit as needed (local path config)

dataset setup

download / create dataset in the path specified in .env
- the folder structure should be dataset_path/system_name/game_name/music.flac
edit ./config/dataset/dataset.json as needed
tbd: dataset processing

new model setup

default model config is in ./config/default, create a copy named ./config/my_model_name
edit config files as needed
run create_new_model.py and enter my_model_name when prompted, your new model will be created in $MODELS_PATH/my_model_name

model training

the model creation process will generate a shell script for training any relevant modules in $MODELS_PATH/my_model_name/training/
edit the training and accelerate config files as needed and then run the shell script to begin training

sampling / testing

run sample.py to launch an interactive local web ui
sampling tests / presets are in $CONFIG_PATH/sampling
the vae module can be sampled / tested by running src/tests/vae.py, config for the vae test is in $CONFIG_PATH/tests/vae_test.json