How to create a tf.Variable with dynamic batch size that need to be given to conv block

I would like to create a tf.Variable with shape=[batch_size,1,512,1] # None,1,512,1.

so that i could pass the variable to convolution layer
Note: i dont want to initialize the batch size before it should take while running

please help me …

Im not sure you can do that because the tf variable itself has to have a determined size, its just a matrix for the model it has no idea that is a batch dimension or not.

And consequently the layers have predefined shapes, inputs and outputs.