Fetch the repository succeeded.
2732
Modified Kernel Primitive API and elementwise for xpu2
Merged
PR types
Others
PR changes
OPs
Describe
Modified Kernel Primitive API and elementwise for xpu2.
PR-CI-OP-benchmark 执行错误的case 如下:
由于代码修改不涉及softmax,重新rerun导致运行超时,因此本地benchmark 验证结果如下:
op name | diff | speed develop (us) | speed new (us) | speed up |
---|---|---|---|---|
softmax_0_forward | 0.00E+00 | 3.097 | 3.107 | 0.32% |
softmax_4_forward | 0.00E+00 | 5.103 | 5.103 | 0.00% |
softmax_0_backward | 5.61E-10 | 3.151 | 3.127 | -0.76% |
softmax_4_backward | 1.63E-08 | 3.856 | 3.842 | -0.36% |