fix likelihood reg v2

sim_transfer/models/bnn_fsvgd.py

@@ -60,7 +60,7 @@ def _neg_log_posterior(self, pred_raw: jnp.ndarray, likelihood_std: jnp.array, x
         neg_log_prior = - self._gp_prior_log_prob(x_stacked, pred_raw, eps=1e-3)
         if self.likelihood_reg > 0:
             likelihood_penalty = self.likelihood_reg * self._likelihood_prior_logprob(jnp.log(likelihood_std))
-            neg_log_prior -= (num_train_points * self.likelihood_exponent) * likelihood_penalty
+            neg_log_prior -= likelihood_penalty
         neg_log_post = nll + neg_log_prior
         stats = OrderedDict(train_nll_loss=nll, neg_log_prior=neg_log_prior)
         if self.learn_likelihood_std: