Add 3.14 support (#830)

Author: oulgen

.github/matrix.json

@@ -9,6 +9,15 @@
       "container-options": "--gpus all",
       "alias": "a10g"
     },
+    {
+      "runner": "linux.g5.4xlarge.nvidia.gpu",
+      "python-version": "3.14",
+      "ref-eager": false,
+      "image": "nvidia/cuda:12.8.1-devel-ubuntu24.04",
+      "runtime-version": "cu128",
+      "container-options": "--gpus all",
+      "alias": "a10g"
+    },
     {
       "runner": "linux.g5.4xlarge.nvidia.gpu",
       "python-version": "3.12",

helion/_testing.py

@@ -80,6 +80,11 @@ def skipIfLowVRAM(
     return unittest.skipIf(os.environ.get("HELION_DEV_LOW_VRAM", "0") == "1", reason)
 
 
+def skipIfPy314(reason: str) -> Callable[[Callable], Callable]:
+    """Skip test if running on Python 3.14"""
+    return unittest.skipIf(sys.version_info >= (3, 14), reason)
+
+
 @contextlib.contextmanager
 def track_run_ref_calls() -> Generator[list[int], None, None]:
     """Context manager that tracks BoundKernel.run_ref calls.

test/test_torch_compile.py

@@ -8,12 +8,14 @@
 from helion._testing import DEVICE
 from helion._testing import RefEagerTestBase
 from helion._testing import TestCase
+from helion._testing import skipIfPy314
 from helion._testing import skipIfRefEager
 import helion.language as hl
 
 
 class TestTorchCompile(RefEagerTestBase, TestCase):
     @skipIfRefEager("does not work with ref eager")
+    @skipIfPy314("torch.compile not yet supported on Python 3.14")
     def test_add_kernel(self):
         @helion.kernel(config=helion.Config(block_sizes=[1, 2]))
         def add(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:

test/test_views.py

@@ -9,6 +9,7 @@
 from helion._testing import RefEagerTestBase
 from helion._testing import TestCase
 from helion._testing import code_and_output
+from helion._testing import skipIfPy314
 from helion._testing import skipIfRocm
 import helion.language as hl
 
@@ -315,6 +316,7 @@ def test_stack_non_power_of_2_kernel(
         torch.testing.assert_close(result, expected, rtol=1e-5, atol=1e-5)
         self.assertExpectedJournal(code)
 
+    @skipIfPy314("torch.compile not yet supported on Python 3.14")
     def test_stack_dim0(self):
         @helion.kernel(use_default_config=True, static_shapes=True)
         def test_stack_dim0_kernel(