diff --git a/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py b/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py
index 4cd663a34..4e32e2258 100644
--- a/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py
+++ b/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py
@@ -2116,7 +2116,7 @@ def __init__(
                 non_critical_for_restore=True,
             )
         self.prior_running_mean_momentum = config.typed_value("prior_running_mean_momentum", None)
-        self.prior_running_mean = None
+        self.prior_running_mean = None  # in std prob, if set
         if self.prior_running_mean_momentum is not None:
             self.prior_running_mean = rf.Parameter(
                 [self.wb_target_dim], auxiliary=True, initial=1.0 / self.wb_target_dim.dimension
@@ -2150,7 +2150,7 @@ def __init__(
         self.log_prob_normed_grad_exclude_blank = config.bool(
             "log_prob_normed_grad_exclude_blank", self.out_blank_separated
         )
-        self.grad_prior = None
+        self.grad_prior = None  # in std prob, if set
         if (
             self.log_prob_normed_grad_opts
             and self.log_prob_normed_grad_opts.get("prior_running_mean_momentum") is not None
@@ -2332,12 +2332,18 @@ def _update_running_stats():
         log_probs = log_probs_am * self.ctc_am_scale
         if self.ctc_prior_scale:
             if self.ctc_prior_type == "batch":
+                # Warning: this is sum, but we want mean!
                 log_prob_prior = rf.reduce_logsumexp(
                     log_probs_am, axis=[dim for dim in log_probs_am.dims if dim != self.wb_target_dim]
                 )
                 assert log_prob_prior.dims == (self.wb_target_dim,)
+            elif self.ctc_prior_type == "batch_fixed":
+                log_prob_prior = rf.reduce_logmeanexp(
+                    log_probs_am, axis=[dim for dim in log_probs_am.dims if dim != self.wb_target_dim]
+                )
+                assert log_prob_prior.dims == (self.wb_target_dim,)
             elif self.ctc_prior_type == "seq":
-                log_prob_prior = rf.reduce_logsumexp(
+                log_prob_prior = rf.reduce_logmeanexp(
                     log_probs_am, axis=[dim for dim in log_probs_am.dims if dim not in (batch_dim, self.wb_target_dim)]
                 )
                 assert log_prob_prior.dims_set == {batch_dim, self.wb_target_dim}