Add new_empty (#8087)

Co-authored-by: Tracy Chen <[email protected]>

Author: tracych477

experimental/torch_xla2/test/test_ops.py

@@ -75,7 +75,6 @@
     "multinomial",
     "mvlgamma",
     "nanmedian",
-    "new_empty",
     "new_empty_strided",
     "nextafter",
     "nn.functional.adaptive_avg_pool3d",
@@ -167,6 +166,7 @@
   'empty_permuted',
   'empty_strided',
   'bernoulli',
+  "new_empty",
   'randint_like',
   'randn',
   'randn_like',

experimental/torch_xla2/torch_xla2/ops/jaten.py

@@ -474,8 +474,6 @@ def _aten__to_copy(self, **kwargs):
   return jnp.copy(self)
 
 
-
-
 @op(torch.ops.aten.empty)
 @op_base.convert_dtype()
 def _aten_empty(size: Sequence[int], *, dtype=None, **kwargs):
@@ -3930,6 +3928,11 @@ def _aten_flatten(x, start_dim=0, end_dim=-1):
   return jnp.reshape(x, new_shape)
 
 
+@op(torch.ops.aten.new_empty)
+def _new_empty(self, size, **kwargs):
+  return jnp.empty(size)
+
+
 @op(torch.ops.aten.new_empty_strided)
 def _new_empty_strided(self, size, stride, **kwargs):
   return jnp.empty(size)