FP16へ autocast されるCUDA Ops:
- Conv系
- Conv系
conv1d
conv2d
conv3d
- ConvT系
conv_transpose1d
conv_transpose2d
conv_transpose3d
- Conv系
- RNN系
RNNCell
LSTMCell
GRUCell
- Linear系
linear
matmul
__matmul__
chain_matmul
baddbmm
addbmm
addmm
addmv
addr
bmm
mm
mv
multi_dot
- Activation系
prelu