73 lines
2.2 KiB
Python
73 lines
2.2 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
import torch
|
|
from torch import Tensor
|
|
|
|
from vllm import ir
|
|
from vllm.platforms import current_platform
|
|
|
|
current_platform.import_kernels()
|
|
|
|
|
|
def is_xpu_kernels_found() -> bool:
|
|
from importlib.util import find_spec
|
|
|
|
return find_spec("vllm_xpu_kernels") is not None
|
|
|
|
|
|
XPU_KERNELS_SUPPORTED = is_xpu_kernels_found()
|
|
"""Kernels in this file are supported if vLLM XPU kernels are installed."""
|
|
|
|
rms_no_var = lambda x, weight, epsilon, variance_size=None: variance_size is None and (
|
|
weight is None or weight.dtype == x.dtype
|
|
)
|
|
|
|
|
|
@ir.ops.rms_norm.register_impl(
|
|
"xpu_kernels", supports_args=rms_no_var, supported=XPU_KERNELS_SUPPORTED
|
|
)
|
|
def rms_norm(
|
|
x: Tensor, weight: Tensor | None, epsilon: float, variance_size: int | None = None
|
|
) -> Tensor:
|
|
assert variance_size is None
|
|
if weight is None:
|
|
# Weightless _C ops are CUDA-only; native skips the multiply on XPU.
|
|
return ir.ops.rms_norm.impls["native"].impl_fn(
|
|
x, weight, epsilon, variance_size
|
|
)
|
|
output = torch.empty(x.shape, device=x.device, dtype=x.dtype)
|
|
torch.ops._C.rms_norm(output, x, weight, epsilon)
|
|
return output
|
|
|
|
|
|
rms_add_no_var_size = (
|
|
lambda x, x_residual, weight, epsilon, variance_size=None: variance_size is None
|
|
and (weight is None or weight.dtype == x.dtype)
|
|
)
|
|
|
|
|
|
@ir.ops.fused_add_rms_norm.register_impl(
|
|
"xpu_kernels",
|
|
supports_args=rms_add_no_var_size,
|
|
supported=XPU_KERNELS_SUPPORTED,
|
|
inplace=True,
|
|
)
|
|
def fused_add_rms_norm(
|
|
x: Tensor,
|
|
x_residual: Tensor,
|
|
weight: Tensor | None,
|
|
epsilon: float,
|
|
variance_size: int | None = None,
|
|
) -> tuple[Tensor, Tensor]:
|
|
assert variance_size is None
|
|
if weight is None:
|
|
# Weightless _C ops are CUDA-only; native skips the multiply on XPU.
|
|
output, residual = ir.ops.fused_add_rms_norm.impls["native"].impl_fn(
|
|
x, x_residual, weight, epsilon, variance_size
|
|
)
|
|
x.copy_(output)
|
|
x_residual.copy_(residual)
|
|
return x, x_residual
|
|
torch.ops._C.fused_add_rms_norm(x, x_residual, weight, epsilon)
|
|
return x, x_residual
|