[XPU][UT]Fix xpu pass_config.fuse_norm_quant assert issue (#46804)
Signed-off-by: Lai, Yejing <yejing.lai@intel.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
Yejing Lai committed
Jun 30, 2026 at 12:13 UTC
ae2c4f3db713242b173548bc46f1b74b4dbb4c08
2 files changed
+7
-3
tests/test_config.py
+3
-1
@@ -1292,7 +1292,9 @@ def test_vllm_config_explicit_overrides():
1292
compilation_config=compilation_config,
1293
)
1294
assert config.compilation_config.cudagraph_mode == CUDAGraphMode.NONE
1295
- assert config.compilation_config.pass_config.enable_qk_norm_rope_fusion is True
1295
+ assert config.compilation_config.pass_config.enable_qk_norm_rope_fusion is (
1296
+ current_platform.is_cuda_alike() or current_platform.is_xpu()
1297
+ )
1298
# Mode should still use default for O2
1299
assert config.compilation_config.mode == CompilationMode.VLLM_COMPILE
1300
vllm/config/compilation.py
+4
-2
@@ -262,10 +262,12 @@ class PassConfig:
262
"Fusion enabled but reshape elimination disabled. "
263
"RMSNorm + padding fusion might not work"
264
)
265
- if self.enable_qk_norm_rope_fusion and not current_platform.is_cuda_alike():
265
+ if self.enable_qk_norm_rope_fusion and not (
266
+ current_platform.is_cuda_alike() or current_platform.is_xpu()
267
+ ):
268
logger.warning_once(
269
"QK Norm + RoPE fusion enabled but the current platform is not "
268
- "CUDA or ROCm. The fusion will be disabled."
270
+ "CUDA, ROCm or XPU. The fusion will be disabled."
271
)
272
self.enable_qk_norm_rope_fusion = False
273
if self.fuse_act_padding and not current_platform.is_rocm():