[Kernel][Helion][1/N] Add Helion kernel for rms_norm_per_block_quant (#36895)

Signed-off-by: Sean Chen <seachen@redhat.com>
Co-authored-by: Yanan Cao <gmagogsfm@gmail.com>
This commit is contained in:
Xiaohong (Sean) Chen
2026-06-16 22:09:52 +08:00
committed by GitHub
co-authored by Yanan Cao
parent bf5149b516
commit ce3ef17bec
5 changed files with 8573 additions and 0 deletions
+15
View File
@@ -2,6 +2,8 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Utility functions for Helion kernel management."""
import torch
from vllm.logger import init_logger
from vllm.platforms import current_platform
@@ -78,3 +80,16 @@ def canonicalize_gpu_name(name: str) -> str:
def get_canonical_gpu_name(device_id: int | None = None) -> str:
return canonicalize_gpu_name(get_gpu_name(device_id))
def get_fp8_dtype() -> torch.dtype:
return current_platform.fp8_dtype()
def get_int8_min_max() -> tuple[int, int]:
qtype_traits = torch.iinfo(torch.int8)
return qtype_traits.min, qtype_traits.max
def get_int8_min_scaling_factor() -> float:
return torch.finfo(torch.float32).eps