Transformers applied to RL tasks

Transformers trained with reinforce and V-MPO

VMPO

GTrXL transformer

Run tests/gtrxl_test.py with --state_rep=gtrxl

CoBERL

Run tests/gtrxl_test.py with --state_rep=coberl to run the architecture with CoBERL in the place of GTrXL in architecture

Reinforce

To run the GTrXL as a policy-estimating network, run pg.py with bool Trans=True within. To evaluate the common MLP estimation, set this bool Trans=False.