v4.1 release update v2. (#2481)

This commit is contained in:
Junkai-Wu
2025-07-22 10:03:55 +08:00
committed by GitHub
parent 9baa06dd57
commit fd6cfe1ed0
179 changed files with 7878 additions and 1286 deletions

View File

@@ -36,7 +36,7 @@
APIs on NVIDIA Blackwell SM100 architecture.
The basic computation logic of dgrad convolution kernel is, take 3D convolution as an example:
Xformed Actication (NZPQK) * Weight/Filter (KTRSC) = Activation (NDHWC)
Xformed Activation (NZPQK) * Weight/Filter (KTRSC) = Activation (NDHWC)
where in terms of GEMM perspective,
Matrix A = Xformed Activation, Matrix B = Weight/Filter, Matrix C = Activation

View File

@@ -36,7 +36,7 @@
APIs on NVIDIA Blackwell SM100 architecture.
The basic computation logic of fprop convolution kernel is, take 3D convolution as an example:
Activation (NDHWC) * Weight/Filter (KTRSC) = Xformed Actication (NZPQK)
Activation (NDHWC) * Weight/Filter (KTRSC) = Xformed Activation (NZPQK)
where in terms of GEMM perspective,
Matrix A = Activation, Matrix B = Weight/Filter, Matrix C = Xformed Activation

View File

@@ -36,7 +36,7 @@
APIs on NVIDIA Blackwell SM100 architecture.
The basic computation logic of wgrad convolution kernel is, take 3D convolution as an example:
Xformed Actication (NZPQK) * Activation (NDHWC) = Weight/Filter (KTRSC)
Xformed Activation (NZPQK) * Activation (NDHWC) = Weight/Filter (KTRSC)
where in terms of GEMM perspective,
Matrix A = Xformed Activation, Matrix B = Activation, Matrix C = Weight/Filter