chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[kernel slimming] Move fast_hadamard_transform to jit_kernel (#18475)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2026-02-14 23:06:21 +08:00
committed by GitHub
parent ae95869292
commit c29394e3c8
17 changed files with 2170 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
sgl-kernel/python/sgl_kernel/__init__.py
View File

@@ -65,13 +65,6 @@ from sgl_kernel.gemm import (
silu_and_mul_scaled_fp4_grouped_quant,
)
from sgl_kernel.grammar import apply_token_bitmask_inplace_cuda
from sgl_kernel.hadamard import (
hadamard_transform,
hadamard_transform_12n,
hadamard_transform_20n,
hadamard_transform_28n,
hadamard_transform_40n,
)
from sgl_kernel.kvcacheio import (
transfer_kv_all_layer,
transfer_kv_all_layer_mla,

21
sgl-kernel/python/sgl_kernel/hadamard.py
View File

@@ -1,21 +0,0 @@
import torch
def hadamard_transform(x: torch.Tensor, scale: float = 1.0) -> torch.Tensor:
return torch.ops.sgl_kernel.fast_hadamard_transform.default(x, scale)
def hadamard_transform_12n(x: torch.Tensor, scale: float = 1.0) -> torch.Tensor:
return torch.ops.sgl_kernel.fast_hadamard_transform_12N.default(x, scale)
def hadamard_transform_20n(x: torch.Tensor, scale: float = 1.0) -> torch.Tensor:
return torch.ops.sgl_kernel.fast_hadamard_transform_20N.default(x, scale)
def hadamard_transform_28n(x: torch.Tensor, scale: float = 1.0) -> torch.Tensor:
return torch.ops.sgl_kernel.fast_hadamard_transform_28N.default(x, scale)
def hadamard_transform_40n(x: torch.Tensor, scale: float = 1.0) -> torch.Tensor:
return torch.ops.sgl_kernel.fast_hadamard_transform_40N.default(x, scale)