diff --git a/src/models/msa_module.py b/src/models/msa_module.py
index 50ece8b..a847eb2 100644
--- a/src/models/msa_module.py
+++ b/src/models/msa_module.py
@@ -26,7 +26,7 @@
 from src.models.components.outer_product_mean import OuterProductMean
 from src.models.components.transition import Transition
 from src.models.components.dropout import DropoutRowwise
-from src.utils.tensor_utils import add
+from src.utils.tensor_utils import add, flatten_final_dims
 from functools import partial
 from typing import Dict
 from src.utils.checkpointing import checkpoint_blocks, get_checkpoint_fn
@@ -121,9 +121,7 @@ def forward(
         o = g * torch.sum(v * weights, dim=-3)  # (*, seq, res, heads, c_hidden)
 
         # Output projection
-        output = self.output_proj(
-            o.reshape((m.shape[:-1] + (self.c_hidden * self.no_heads,)))  # (*, seq, res, c_hidden * heads)
-        )
+        output = self.output_proj(flatten_final_dims(o, 2))  # (*, seq, res, c_hidden * heads)
         return output
 
 
diff --git a/src/utils/geometry/alignment.py b/src/utils/geometry/alignment.py
index aa2d731..6e29d82 100644
--- a/src/utils/geometry/alignment.py
+++ b/src/utils/geometry/alignment.py
@@ -22,7 +22,8 @@ def weighted_rigid_align(
         x: Vec3Array,  # (bs, n_atoms)
         x_gt: Vec3Array,  # (bs, n_atoms)
         weights: torch.Tensor,  # (bs, n_atoms)
-        mask: torch.Tensor = None  # (bs, n_atoms)
+        mask: torch.Tensor = None,  # (bs, n_atoms)
+        eps: float = 1e-5
 ) -> Vec3Array:
     """Performs a weighted alignment of x to x_gt. Warning: ground truth here only refers to the structure
     not being moved, not to be confused with ground truth during training."""
@@ -32,8 +33,8 @@ def weighted_rigid_align(
         torch.set_float32_matmul_precision("highest")
 
         # Mean-centre positions
-        mu = (x * weights).mean(dim=-1, keepdim=True) / weights.mean(dim=-1, keepdim=True)
-        mu_gt = (x_gt * weights).mean(dim=-1, keepdim=True) / weights.mean(dim=-1, keepdim=True)
+        mu = (x * weights).sum(dim=-1, keepdim=True) / (weights.sum(dim=-1, keepdim=True) + eps)
+        mu_gt = (x_gt * weights).sum(dim=-1, keepdim=True) / (weights.sum(dim=-1, keepdim=True) + eps)
         x -= mu  # Vec3Array of shape (*, n_atoms)
         x_gt -= mu_gt