How to train or compress large models? Today the data is really big and the models are really heavy. So, it is essential to know hints how to train such models. Here some interesting recent papers about this topic. How to Train Really Large Models on Many GPUs? A tutorial, how you can parallel your data or your model to make it possible to train on GPUs. Compressing Large-Scale Transformer-Based Models: A Case Study on BERT The paper specifically about BERT and how such hints as, for example, pruning or quantization, can be applied. The topic of efficient modern models training is still challenging and personally I want to know more about it. So, if you have any your personal favourite hints or interesting sources/tutorials that you use, you are very welcome to share!🤗