Large Model Support (LMS) is a Python library recently launched by IBM. This library has been ideated in order to train large Deep Learning models which can’t fit in GPU memory. In fact, GPUs have generally smaller memory space compared to Central Processing Units (CPUs)

Above Signal Flow provides guide line to handle "Traning of Large Model". For example, it is easy to cut down data set size and  train DL Networks Model.  Or cut down DL networks size and train.  But these two methods appears to be leading to high amount of in accuracy.  Kears appears to be given library to perform training of Large Model by using Graph and its pre-processing.  

Large Model Reference from IBM and Keras

LLMs Introduction to Hugging Face

Sequence in Modelling