Fixed mask sharding if inputs is sharded

flax/nnx/nn/stochastic.py

@@ -150,7 +150,9 @@ def __call__(
     broadcast_shape = list(inputs.shape)
     for dim in self.broadcast_dims:
       broadcast_shape[dim] = 1
-    mask = random.bernoulli(key, p=keep_prob, shape=broadcast_shape)
+    mask = random.bernoulli(
+      key, p=keep_prob, shape=broadcast_shape, out_sharding=jax.typeof(inputs).sharding
+    )
     mask = jnp.broadcast_to(mask, inputs.shape)
     return lax.select(mask, inputs / keep_prob, jnp.zeros_like(inputs))
 

tests/nnx/spmd_test.py

@@ -243,6 +243,24 @@ def test_out_sharding_embed_attend(self):
       assert 'float32[2@X,10]' in str(jax.typeof(layer.attend(sharded_array)))
       assert 'float32[2@X,10@Y]' in str(jax.typeof(layer.attend(sharded_array, out_sharding=P("X", "Y"))))
 
+  def test_out_sharding_dropout(self):
+    mesh = jax.make_mesh((2, 2), ("X", "Y"), axis_types=(AxisType.Explicit, AxisType.Explicit))
+    with jax.set_mesh(mesh):
+      replicated_array = jnp.arange(8).reshape(2, 4).astype(jnp.float32)
+      sharded_array = reshard(replicated_array, P("X", None))
+      layers = [
+        nnx.Dropout(rate=0.5, rngs=nnx.Rngs(0)),
+        nnx.Dropout(rate=0.5, broadcast_dims=(1,), rngs=nnx.Rngs(0)),
+      ]
+      for layer in layers:
+        assert 'float32[2@X,4]' in str(jax.typeof(layer(sharded_array)))
+
+        @jax.jit
+        def func(x, rngs):
+          return layer(x, rngs=rngs)
+
+        assert 'float32[2@X,4]' in str(jax.typeof(func(sharded_array, nnx.Rngs(0))))
+
   @parameterized.product(use_hijax=[True, False])
   def test_logical_rules(self, use_hijax):
     self.enter_context(nnx.var_defaults(hijax=use_hijax))