In the Doc “RECURRENT DQN: TRAINING RECURRENT POLICIES”

I managed to run the code, but during the process, I realized that the maximum STEP for each batch is only 50，
steps: 50, loss_val: 0.1930, action_spread: tensor([26, 24], device='cuda:0'): 18%|█▊ | 181450/1000000 [1:54:35<9:08:14, 24.88it/s]
I tried to output it
print(data[ "step_count"])

tensor([[ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 9],
        [10],
        [11],
        [12],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 9]], device='cuda:0')

next output is

tensor([[10],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 9],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 9],
        [10],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 0],
        [ 1],
        [ 2],
        [ 3],
        [ 4],
        [ 5],
        [ 6],
        [ 7],
        [ 8],
        [ 0]], device='cuda:0')

I've tried many times and it's the same pattern, that is to say, the accounting number will start again after each batch. I don't know why.