First, I really recommend to go pro+, I am working on colab for two years now and it is usually sufficient for nlp/vision POCs. Later, as projects get more complicated, it is recommended to move to a larger machine (im a master’s student as well and I work on the university cluster, check if there is one at your university, most likely there is one). Second, try maybe to change your training type, and work with generators instead of loading the whole data into memory at once.