Any suggestions on continous error "ValueError: shape mismatch: value array of shape (62,) could not be broadcast to indexing result of shape (0,)"

Thank you so much for your suggestions. There are indeed solved the problem.

One more quick question:
My current likelihood is Multinomial. When I set the observation matrix 1 * 62, the program took around 4 minutes, but very very slow when observation matrix is 96 * 62. I checked the time needed of codes before pm.sample(), less than 1 s. I wonder, is my data large or the multinomial is inherently slow? Any suggestions on speeding up?