Remove import intel_extension_for_pytorch from fused_softmax.py (#…

…2278) Part of #2147 Perf the same: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/10928161213 vs https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/10912513726; geomean diff: 1% for triton, 2% for xetla One more CI run: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/10928754028 (also looks good) Signed-off-by: Anatoly Myachev <anatoly.myachev@intel.com>
intel · Sep 19, 2024 · e820f6c · e820f6c
1 parent 4d546ad
commit e820f6c
Showing 1 changed file with 0 additions and 3 deletions.
diff --git a/benchmarks/triton_kernels_benchmark/fused_softmax.py b/benchmarks/triton_kernels_benchmark/fused_softmax.py
@@ -15,9 +15,6 @@
 import triton_kernels_benchmark as benchmark_suit
 import xetla_kernel
 
-if benchmark_suit.USE_IPEX_OPTION:
-    import intel_extension_for_pytorch  # type: ignore # noqa: F401
-
 
 @torch.jit.script
 def naive_softmax(x):