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

[XPU]REPO cache dtype xpu align with cuda (#31140)

Browse Source
This commit is contained in:
Swift.Sun
2026-07-16 10:30:01 +08:00
committed by GitHub
parent 3a8ddd3fd0
commit f28ce5c420
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/rotary_embedding/base.py
+1 -1
View File
@@ -95,7 +95,7 @@ class RotaryEmbedding(MultiPlatformOp):
cache = self._compute_cos_sin_cache() cache = self._compute_cos_sin_cache()
# NOTE(ByronHsu): cache needs to be in FP32 for numerical stability. # NOTE(ByronHsu): cache needs to be in FP32 for numerical stability.
if not (_is_cuda or envs.SGLANG_ROPE_CACHE_FP32.get()): if not (_is_cuda or _is_xpu or envs.SGLANG_ROPE_CACHE_FP32.get()):
cache = cache.to(dtype) cache = cache.to(dtype)
if ( if (