allow for rotary embeddings to be constructed outside the attention l…

…ayers and passed in

setup.py

@@ -3,7 +3,7 @@
 setup(
   name = 'x-transformers',
   packages = find_packages(exclude=['examples']),
-  version = '1.25.2',
+  version = '1.25.3',
   license='MIT',
   description = 'X-Transformers - Pytorch',
   author = 'Phil Wang',

x_transformers/x_transformers.py

@@ -1191,7 +1191,8 @@ def forward(
         seq_start_pos: Optional[Tensor] = None,
         cache: Optional[LayerIntermediates] = None,
         cache_age = 1,
-        return_hiddens = False
+        return_hiddens = False,
+        rotary_pos_emb = None
     ):
         assert not (self.cross_attend ^ exists(context)), 'context must be passed in if cross_attend is set to True'
 
@@ -1219,9 +1220,7 @@ def forward(
 
         # rotary positions
 
-        rotary_pos_emb = None
-
-        if exists(self.rotary_pos_emb):
+        if not exists(rotary_pos_emb) and exists(self.rotary_pos_emb):
             max_rotary_emb_length = max(list(map(lambda m: (m.shape[1] if exists(m) else 0) + x.shape[1], mems)))
             rotary_pos_emb = self.rotary_pos_emb(max_rotary_emb_length)