minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[Intel GPU] DeepSeek V4 7/N: Support fused_rope_inplace on XPU using triton (#27915)

Browse Source 
Signed-off-by: P V R K Jyothendra Varma <polisettyvarma@gmail.com>
Co-authored-by: Rahul Vijayaraghavan <rahul.vijayaraghavan@intel.com>
Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
This commit is contained in:
Polisetty V R K Jyothendra Varma
2026-07-02 13:22:20 +08:00
committed by GitHub
co-authored by Rahul Vijayaraghavan Ma Mingfei
parent cb06c4e6ce
commit 790156c98f
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/jit_kernel/dsv4/elementwise.py
+3 -2
View File
@@ -8,11 +8,12 @@ from sglang.jit_kernel.utils import (
load_jit, load_jit,
make_cpp_args, make_cpp_args,
) )
from sglang.srt.utils import is_hip from sglang.srt.utils import is_hip, is_xpu
from .utils import make_name from .utils import make_name
_is_hip = is_hip() _is_hip = is_hip()
_is_xpu = is_xpu()
@cache_once @cache_once
@@ -121,7 +122,7 @@ def fused_rope_inplace(
positions: [batch_size] int32 or int64, indices into freqs_cis positions: [batch_size] int32 or int64, indices into freqs_cis
inverse: if True, apply inverse rotation (conjugate freqs) inverse: if True, apply inverse rotation (conjugate freqs)
""" """
if _is_hip: if _is_hip or _is_xpu:
from sglang.srt.layers.deepseek_v4_rope import apply_rotary_emb_triton from sglang.srt.layers.deepseek_v4_rope import apply_rotary_emb_triton
apply_rotary_emb_triton(q, freqs_cis, positions=positions, inverse=inverse) apply_rotary_emb_triton(q, freqs_cis, positions=positions, inverse=inverse)