x-transformers 2.7.0__tar.gz → 2.7.1__tar.gz

{x_transformers-2.7.0 → x_transformers-2.7.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: x-transformers
-Version: 2.7.0
+Version: 2.7.1
 Summary: X-Transformers
 Project-URL: Homepage, https://pypi.org/project/x-transformers/
 Project-URL: Repository, https://github.com/lucidrains/x-transformers

{x_transformers-2.7.0 → x_transformers-2.7.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "x-transformers"
-version = "2.7.0"
+version = "2.7.1"
 description = "X-Transformers"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }

{x_transformers-2.7.0 → x_transformers-2.7.1}/x_transformers/continuous.py

@@ -241,6 +241,7 @@ class ContinuousAutoregressiveWrapper(Module):
         self,
         net: ContinuousTransformerWrapper,
         loss_fn: Module | None = None,
+        use_l1_loss = False,
         equal_loss_weight_batch = False,  # setting this to True, if the mask is passed in and sequences are variable in length, each sequence will be weighted the same (as opposed to each token)
     ):
         super().__init__()
@@ -250,7 +251,15 @@ class ContinuousAutoregressiveWrapper(Module):
         probabilistic = net.probabilistic
         self.probabilistic = probabilistic
 
-        loss_fn = default(loss_fn, nn.MSELoss(reduction = 'none') if not probabilistic else GaussianNLL())
+        # default loss function
+
+        if not exists(loss_fn):
+            if probabilistic:
+                loss_fn = GaussianNLL()
+            elif use_l1_loss:
+                loss_fn = nn.L1Loss(reduction = 'none')
+            else:
+                loss_fn = nn.MSELoss(reduction = 'none')
 
         self.loss_fn = loss_fn
         self.equal_loss_weight_batch = equal_loss_weight_batch