Problem with sampling from a CompletedDerichlet

am trying to implement LDA algorithm and I have some problems with tracing from CompletedDirichlet distribution .

I am using the following code

import numpy as np
import pymc as pm

K = 2 # number of topics

beta = np.ones(K)
phi = pm.Container([pm.CompletedDirichlet("phi_%s" % k, pm.Dirichlet("pphi_%s" % k, theta=beta)) for k in range(K)])
model = pm.Model([phi])
mcmc = pm.MCMC(model)

My problem is that when I do mcmc.trace('phi_0')[:] i get the same values for each sample

completederichlet: how can I make MCMC to give me different samples?