a bug when using MultivariateStudentTLinearOperator in DistributionLamda during training

I want to build a distribution layer of multivariate students't distribution. when I using tensorflow.distribution.MultivariateStudentTLinearOperator in DistributionLambda, always have an error:

`AttributeError: Exception encountered when calling layer "distribution_lambda_20" (type DistributionLambda).

'Tensor' object has no attribute 'range_dimension'

Call arguments received:
• inputs=['tf.Tensor(shape=(None, 64), dtype=float32)', 'tf.Tensor(shape=(None, 64), dtype=float32)']
• args=<class 'inspect._empty'>
• kwargs={'training': 'None'}`

My code:
`

inputs = tfk.Input(shape=input_shape)

x1= tfkl.Dense(64)(inputs)
x2= tfkl.Dense(64)(inputs)
outputs=tfpl.DistributionLambda(
make_distribution_fn=lambda t: tfd.multivariate_student_t.MultivariateStudentTLinearOperator (df=3,loc=t[0],scale=1e-05+ tf.math.softplus(0.01 *t[1]))
)([x1, x2]) `

This seems like could not receive 'None', but it works when using tfd.StudenT. Is anyone know what is the problem? Any solution? Thank you very much!

Start by running the provided TensorFlow Probability reproduction with DistributionLambda and MultivariateStudentTLinearOperator, comparing it with the working StudentT case. Trace the range_dimension error during training and confirm the issue is resolved when the multivariate distribution layer accepts the shown inputs without raising the AttributeError.