27 #ifndef LBANN_LAYERS_MISC_ARGMAX_HPP_INCLUDED 28 #define LBANN_LAYERS_MISC_ARGMAX_HPP_INCLUDED 40 template <
typename TensorDataType, data_layout Layout, El::Device Device>
44 "argmax layer only supports data parallel layout");
45 static_assert(
Device == El::Device::CPU,
"argmax layer only supports CPU");
54 template <
typename ArchiveT>
59 std::string
get_type()
const override {
return "argmax"; }
77 #ifndef LBANN_ARGMAX_LAYER_INSTANTIATE 79 extern template class argmax_layer<T, \ 80 data_layout::DATA_PARALLEL, \ 83 #define LBANN_INSTANTIATE_CPU_HALF 86 #undef LBANN_INSTANTIATE_CPU_HALF 87 #endif // LBANN_ARGMAX_LAYER_INSTANTIATE 91 #endif // LBANN_LAYERS_MISC_ARGMAX_HPP_INCLUDED argmax_layer(lbann_comm *comm)
void setup_dims() override
Setup tensor dimensions Called by the 'setup' function. If there are any input tensors, the base method sets all uninitialized output tensor dimensions equal to the first input tensor dimensions.
void write_specific_proto(lbann_data::Layer &proto) const final
argmax_layer * copy() const override
Copy function. This function dynamically allocates memory for a layer instance and instantiates a cop...
El::Device get_device_allocation() const override
Get the device allocation for the data tensors. We assume that the decice allocation of the previous ...
constexpr El::Device Device
int get_backprop_requirements() const override
Returns the necessary tensors for computing backpropagation.
void serialize(ArchiveT &ar)
data_layout get_data_layout() const override
Get data layout of the data tensors. We assume that the data layouts of the previous activations...
data_layout
Data layout that is optimized for different modes of parallelism.
std::string get_type() const override
Get the layer type's name.
void fp_compute() override
Apply layer operation. Called by the 'forward_prop' function. Given the input tensors, the output tensors are populated with computed values.
friend class cereal::access
Get index of maximum-value tensor entry.
bool can_run_inplace() const override
If True, the computation can run in-place (feeding each input activations tensor as the corresponding...