ValueError: at site "N_X", invalid log_prob shape Expected [], actual [4096]

4

@jpchen these were generated by printing .shape for each noise variable:

----- guide -------- 
N_X torch.Size([4096]) 
N_Z torch.Size([50]) 
N_Y_1 torch.Size([3]) 
N_Y_2 torch.Size([6]) 
N_Y_3 torch.Size([40]) 
N_Y_4 torch.Size([32]) 
N_Y_5 torch.Size([32]) 
----- model ----- 
N_X torch.Size([4096]) 
N_Z torch.Size([50]) 
N_Y_1 torch.Size([3]) 
N_Y_2 torch.Size([6]) 
N_Y_3 torch.Size([40]) 
N_Y_4 torch.Size([32]) 
N_Y_5 torch.Size([32])

We get the same results if we print the batch_size of each underlying distributions