How to implement adversarial training?

hi guys, i was wondering how can implement adversarial training in my mlp when i have a distributional lambda layer t hat model a gaussian as last layer
tfp.layers.DistributionLambda(normal_sp, name='normal_sp')( params_mc )

with normal_sp defined as

def normal_sp(params): 

        #return tfd.Normal(loc=params[:,0:1], scale=1e-5 +  params[:,1:2]) 

        return tfd.Normal(loc= params[:,0:1], scale=1e-3 +tf.math.softplus(0.005* params[:,1:2])  )

and my loss fucntion is simply

def NLL(y, distr): 
    return -distr.log_prob(y)