SYNTH-GPT2-FIN_conv-10_shift-1_p-0.3_lr-0.001_n_embd-128_n_head-2_n_layer-2_activation_function-relu_global_step=58593.0_train_loss=0.00.ckpt