x-transformers 2.3.22__py3-none-any.whl → 2.3.23__py3-none-any.whl

x_transformers/attend.py

@@ -276,21 +276,22 @@ class Attend(Module):
 
         # torch 2.3 uses new backend and context manager
 
-        if torch_version >= version.parse('2.3'):
-            from torch.nn.attention import SDPBackend
-
-            str_to_backend = dict(
-                enable_flash = SDPBackend.FLASH_ATTENTION,
-                enable_mem_efficient = SDPBackend.EFFICIENT_ATTENTION,
-                enable_math = SDPBackend.MATH,
-                enable_cudnn = SDPBackend.CUDNN_ATTENTION
-            )
+        if self.flash:
+            if torch_version >= version.parse('2.3'):
+                from torch.nn.attention import SDPBackend
 
-            sdpa_backends = [str_to_backend[enable_str] for enable_str, enable in sdp_kwargs.items() if enable]
+                str_to_backend = dict(
+                    enable_flash = SDPBackend.FLASH_ATTENTION,
+                    enable_mem_efficient = SDPBackend.EFFICIENT_ATTENTION,
+                    enable_math = SDPBackend.MATH,
+                    enable_cudnn = SDPBackend.CUDNN_ATTENTION
+                )
 
-            self.sdp_context_manager = partial(torch.nn.attention.sdpa_kernel, sdpa_backends)
-        else:
-            self.sdp_context_manager = partial(torch.backends.cuda.sdp_kernel, **sdp_kwargs)
+                sdpa_backends = [str_to_backend[enable_str] for enable_str, enable in sdp_kwargs.items() if enable]
+
+                self.sdp_context_manager = partial(torch.nn.attention.sdpa_kernel, sdpa_backends)
+            else:
+                self.sdp_context_manager = partial(torch.backends.cuda.sdp_kernel, **sdp_kwargs)
 
     def flash_attn(
         self,

{x_transformers-2.3.22.dist-info → x_transformers-2.3.23.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: x-transformers
-Version: 2.3.22
+Version: 2.3.23
 Summary: X-Transformers
 Project-URL: Homepage, https://pypi.org/project/x-transformers/
 Project-URL: Repository, https://github.com/lucidrains/x-transformers

{x_transformers-2.3.22.dist-info → x_transformers-2.3.23.dist-info}/RECORD

@@ -1,5 +1,5 @@
 x_transformers/__init__.py,sha256=h3I2ejobgEdy8H7NgV-rP8UaBCnd16-MysvDXH9GMEA,985
-x_transformers/attend.py,sha256=fXMuwHuBAFB4f4_U6j5_uVeK7N4cV0PDd6UTqtkjKKM,17333
+x_transformers/attend.py,sha256=Ax34Rw56xXAWfFPqtZ_B8iKEW2EfQdbVoc9uFjfeNjA,17404
 x_transformers/autoregressive_wrapper.py,sha256=BWFaO-3YWzCcEfp-EC1ZkdckqDpPIOQG6_uyyP6AmhM,11753
 x_transformers/belief_state_wrapper.py,sha256=YLUMk6t2MhFBEw5lHDDHJHcoCxTIkHvxTNY__GGZEKU,13374
 x_transformers/continuous.py,sha256=hpb1sSbt3k2LNzzjrjSd8F5xOIbKj7IluV9MBEAFLkw,13031
@@ -11,7 +11,7 @@ x_transformers/nonautoregressive_wrapper.py,sha256=2NU58hYMgn-4Jzg3mie-mXb0XH_dC
 x_transformers/x_transformers.py,sha256=7phSZvP1_SDRIkVMwVR4cz1dFU2UlR2Wf1HJHEQlcQg,116222
 x_transformers/xl_autoregressive_wrapper.py,sha256=CvZMJ6A6PA-Y_bQAhnORwjJBSl6Vjq2IdW5KTdk8NI8,4195
 x_transformers/xval.py,sha256=AwwYUm8yDAtKQyKJDIhYMsiLTJ_skh3scUFMjp5sda8,8597
-x_transformers-2.3.22.dist-info/METADATA,sha256=_8m8ftpHRKjbEUDuoeYPcVh4yan1FxNRj3seJwiZzl8,89897
-x_transformers-2.3.22.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-x_transformers-2.3.22.dist-info/licenses/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-2.3.22.dist-info/RECORD,,
+x_transformers-2.3.23.dist-info/METADATA,sha256=xRMZP1TSYdcbc0F5GX-WcaHhAbQPdGeFIbjHBZYG9_0,89897
+x_transformers-2.3.23.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+x_transformers-2.3.23.dist-info/licenses/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-2.3.23.dist-info/RECORD,,