leepanic/ecot-libero-10-r32

training params: order, bs=4x4, reasoning_dropout_prob=0.05 | batch SR 69%, async SR 69%

training params: order, bs=1x8, reasoning_dropout_prob=0.05 | batch SR 70%

training params: order, reasoning_dropout_prob=0.05 | batch SR 42%, async SR 48%

initial commit