Increase tolerance for glu Half/BFloat16 test (pytorch#7983)

Unbreaks internal tests.

Author: swolchok

kernels/test/op_glu_test.cpp

@@ -28,6 +28,19 @@ class OpGluOutTest : public OperatorTest {
     return torch::executor::aten::glu_outf(context_, self, dim, out);
   }
 
+  template <ScalarType DTYPE>
+  void expect_tensor_close(Tensor actual, Tensor expected) {
+    if (DTYPE == ScalarType::Half || DTYPE == ScalarType::BFloat16) {
+      EXPECT_TENSOR_CLOSE_WITH_TOL(
+          actual,
+          expected,
+          1e-2,
+          executorch::runtime::testing::internal::kDefaultAtol);
+    } else {
+      EXPECT_TENSOR_CLOSE(actual, expected);
+    }
+  }
+
   // Common testing for glu operator
   template <ScalarType DTYPE, ScalarType OUT_DTYPE>
   void test_glu_out() {
@@ -41,14 +54,14 @@ class OpGluOutTest : public OperatorTest {
     Tensor in = tf.ones(sizes);
     Tensor out = tf_out.zeros(out_sizes_1);
     op_glu_out(in, 0, out);
-    EXPECT_TENSOR_CLOSE(
+    expect_tensor_close<DTYPE>(
         out,
         tf_out.make(
             out_sizes_1, /*data=*/{0.731059, 0.731059, 0.731059, 0.731059}));
     const std::vector<int32_t> out_sizes_2 = {4, 1};
     out = tf_out.zeros(out_sizes_2);
     op_glu_out(in, 1, out);
-    EXPECT_TENSOR_CLOSE(
+    expect_tensor_close<DTYPE>(
         out,
         tf_out.make(
             out_sizes_2, /*data=*/{0.731059, 0.731059, 0.731059, 0.731059}));