[Kernel] [Helion] Force disable HOP path due to performance regression (#40171)

Signed-off-by: Yanan Cao <gmagogsfm@gmail.com>
Co-authored-by: Claude Sonnet 4 <noreply@anthropic.com>
This commit is contained in:
Yanan Cao
2026-04-17 17:36:49 -04:00
committed by GitHub
co-authored by Claude Sonnet 4
parent 6ef1efd51f
commit 5cdddddd4a
+3 -2
View File
@@ -53,14 +53,15 @@ if not has_helion():
)
import helion
from helion._compat import requires_torch_version
from helion.autotuner.base_search import BaseAutotuner
from helion.runtime.config import Config
from helion.runtime.settings import default_autotuner_fn
# TODO(gmagogsfm): Remove CustomOp fallback path (_get_or_register_custom_op,
# vllm_helion_lib, direct_register_custom_op) once vLLM requires PyTorch >= 2.11.
_HOP_AVAILABLE = requires_torch_version("2.11")
# FIXME(gmagogsfm): Re-enable HOP path once performance regression is fixed.
# _HOP_AVAILABLE = requires_torch_version("2.11")
_HOP_AVAILABLE = False
if _HOP_AVAILABLE:
from helion._compat import supports_torch_compile_fusion