minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Intel GPU] fix device in DeepseekScalingRotaryEmbedding to run DeepSeek-V2-Lite BF16 on XPU (#10021)

Browse Source 
Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
This commit is contained in:
Polisetty V R K Jyothendra Varma
2026-01-29 21:21:38 -08:00
committed by GitHub
co-authored by Ma Mingfei
parent 0e4d9ddbd6
commit 858dc80aff
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/rotary_embedding.py
+3 -2
View File
@@ -17,6 +17,7 @@ from sglang.srt.utils import (
cpu_has_amx_support, cpu_has_amx_support,
get_bool_env_var, get_bool_env_var,
get_compiler_backend, get_compiler_backend,
get_device,
is_cpu, is_cpu,
is_cuda, is_cuda,
is_hip, is_hip,
@@ -822,7 +823,7 @@ class DeepseekScalingRotaryEmbedding(RotaryEmbedding):
beta_slow: int = 1, beta_slow: int = 1,
mscale: float = 1, mscale: float = 1,
mscale_all_dim: float = 0, mscale_all_dim: float = 0,
device: Optional[str] = "cuda" if not _is_npu else "npu", device: Optional[str] = None,
) -> None: ) -> None:
self.scaling_factor = scaling_factor self.scaling_factor = scaling_factor
self.extrapolation_factor = extrapolation_factor self.extrapolation_factor = extrapolation_factor
@@ -839,7 +840,7 @@ class DeepseekScalingRotaryEmbedding(RotaryEmbedding):
self.sin_cached_total = None self.sin_cached_total = None
self.cos_cached = None self.cos_cached = None
self.sin_cached = None self.sin_cached = None
self.device = device self.device = device if device is not None else get_device()
super().__init__( super().__init__(
head_size, rotary_dim, max_position_embeddings, base, is_neox_style, dtype head_size, rotary_dim, max_position_embeddings, base, is_neox_style, dtype
) )