forked from tinygrad/tinygrad
don't require contiguous after fuse (#10074)
This commit is contained in:
+1
-1
@@ -2750,7 +2750,7 @@ class Tensor(SimpleMathTrait):
|
||||
Useful for single kernel softmax and flash attention.
|
||||
Careful, this can break codegen or make kernels really slow.
|
||||
"""
|
||||
return self._apply_uop(UOp.fuse).contiguous()
|
||||
return self._apply_uop(UOp.fuse)
|
||||
|
||||
def contiguous_backward(self) -> Tensor:
|
||||
"""
|
||||
|
||||
Reference in New Issue
Block a user