0.1.9 mps now works but only to compute the ESM2 embeddings, removed redundancy in predict_string.py

GPU running fixed

bugfix in params.device

output contains gene names instead of STRING IDs

Datasets are now taken from STRING v12.0
model.py - batch_size and lr defaults are 32 and 1e-4 respectively
commands are edited with respect to new version of STRING

0.1.3 fixed max_len bug in dataset.py (was not transfered first due to a wrong final version in previous repository)

Comments, arguments and outdated code have to be cleaned, the version still needs some testing

Working commands: predict, test, string_dataset_create
train and predict_string are still in progress