Train_batch_size hindering observation space

I am working with offline tabular data with 34 columns of continuous values. I have been using marwil so far, but when I switch to CRR I get errors from my input observation space due to the batch size in training.

For example this code for marwil will run fine using this input into the config,
config = MARWILConfig().training().framework("torch").environment(observation_space = Dict({ 'obs': Box(low = -10000.0, high = 100000.0, shape = (34,), dtype = np.float32) }), action_space = Discrete(2) )
But applying this same form to CRRConfig will result in an error,
Expected flattened obs shape of [..., 34], got (128,)
Where 128 is the default training batch size. I tried specifying the batch size in the input observation space, but I can not fix this error.
Has anyone experienced this error? Any help is appreciated. Thank you