agaier / Minimal-Multimodel-Training

Minimal example of parallel model training in PyTorch.

Minimal Multimodel Training

A simple reference implementation to train multiple models on one or more GPUs in parallel with PyTorch.

What's inside:

An minimal example of training multiple networks on MNIST in parallel with a multiprocessing Queue structure:

• single_gpu.py parallel training of multiple models with a single GPU
• multi_gpu.py as above but with multiple GPUs
• main.py a simple implementation of population based training, from here.

Requirements

• PyTorch >= 1.0.0