diff --git a/README.md b/README.md index da8d118..0d9535a 100644 --- a/README.md +++ b/README.md @@ -10,6 +10,15 @@ This toolkit is currently under development and aims to focus on biomedical task pip install polus ``` +For training the models in multiGPU scenario, polus leverage the horovod library for dataparalelism, which means that horovod must be +correctly installed to be able to run the training in multiGPUs. + +To easy this process we made available a docker image with all the correct dependencies to perform the multiGPU training in polus. + +``` +docker pull talmeidawastaken/polus +``` + ### Documentation Still in work, but some of it can be already consulted here: https://bioinformatics-ua.github.io/polus/ diff --git a/polus/__init__.py b/polus/__init__.py index 8393b1c..0d3751a 100644 --- a/polus/__init__.py +++ b/polus/__init__.py @@ -75,7 +75,7 @@ def __init__(self): else: if hvd.size() <= 1: print(f"The script found multiple GPUs and a horovod.tensorlfow installation. However," - f" the only one process was found, please check if you are runing the script with horovodrun.") + f" only one process was initialized, please check if you are runing the script with horovodrun or mpirun.") else: if hvd.local_rank() == 0: print(f"MultiGPU training enabled, using {hvd.size()} processes ")