Prefix location
must be located under /scratch/$USER
because the limit on the number of files in the home directory is only 100k"${SHELL}" <(curl -L micro.mamba.pm/install.sh)
copied from Micromamba Installation
You can install python packages with environment.yaml
micromamba create -y -f environment.yaml
let's start setup!
source setup.sh
You can then train a deep learning model with train.py
and a configuration .yaml file.
For example, you can a Transformer-based network that takes in L1 ParticleFlow candidates and reconstruct MET.
python train.py -c ./config/test-l1pf-transformer-neuron.yaml
If you want to understand how to configure it, I recommend reading Configure hyperparameters from the CLI
Before submitting a job, u need to check available memory.
condor_status -compact -constraint 'Gpus > 0'
You can submit a training job to a GPU cluster using submit.py
submit.py --config ./config/only-high-met.yaml --data ./config/data/delphes/knu/sanity-check.yaml --trainer ./config/trainer/sanity-check.yaml -e sanity-check -m 50GB
condor_q $USER