infini::ops::TopkSoftmax Class Reference

#include <topk_softmax.h>

Inheritance diagram for infini::ops::TopkSoftmax:

Public Member Functions
	TopkSoftmax (const Tensor gating_output, std::optional< Tensor > bias, std::optional< Tensor > is_padding, const bool renormalize, Tensor topk_weights, Tensor topk_indices, Tensor token_expert_indices)
virtual void	operator() (const Tensor gating_output, std::optional< Tensor > bias, std::optional< Tensor > is_padding, const bool renormalize, Tensor topk_weights, Tensor topk_indices, Tensor token_expert_indices) const =0
Public Member Functions inherited from infini::ops::Operator< TopkSoftmax >
void	operator() (const Handle &handle, const Args &... args)
void	operator() (const Args &... args) const
void	operator() (const Handle &handle, const Args &... args)
void	operator() (const Args &... args) const
Public Member Functions inherited from infini::ops::OperatorBase
virtual	~OperatorBase ()=default
virtual std::size_t	workspace_size_in_bytes () const
void	set_handle (const Handle &handle)
void	set_config (const Config &config)
void	set_stream (void *stream)
void	set_workspace (void *workspace)
void	set_workspace_size_in_bytes (std::size_t workspace_size_in_bytes)
virtual	~OperatorBase ()=default
virtual std::size_t	workspace_size_in_bytes () const
void	set_handle (const Handle &handle)
void	set_config (const Config &config)
void	set_stream (void *stream)
void	set_workspace (void *workspace)
void	set_workspace_size_in_bytes (std::size_t workspace_size_in_bytes)

Protected Member Functions
void	ValidateCallMetadata (const Tensor gating_output, std::optional< Tensor > bias, std::optional< Tensor > is_padding, const bool renormalize, Tensor topk_weights, Tensor topk_indices, Tensor token_expert_indices) const

Protected Attributes
Tensor::Size	num_tokens_ {0}
Tensor::Size	num_experts_ {0}
Tensor::Size	topk_ {0}
DataType	input_dtype_
DataType	index_dtype_
bool	renormalize_ {false}
int	device_index_ {0}
Protected Attributes inherited from infini::ops::OperatorBase
std::unique_ptr< Handle >	handle_ptr_
std::unique_ptr< Config >	config_ptr_
void *	stream_ {nullptr}
void *	workspace_ {nullptr}
std::size_t	workspace_size_in_bytes_ {0}

Additional Inherited Members
Static Public Member Functions inherited from infini::ops::Operator< TopkSoftmax >
static void	clear_cache ()
static void	clear_cache ()
static std::unique_ptr< Operator >	Make (const Config &config, const Tensor tensor, Args &&... args)
static std::unique_ptr< Operator >	Make (const Tensor tensor, Args &&... args)
static std::unique_ptr< Operator >	Make (const Config &config, const std::vector< Tensor > tensors, Args &&... args)
static std::unique_ptr< Operator >	Make (const std::vector< Tensor > tensors, Args &&... args)
static std::unique_ptr< Operator >	Make (const Config &config, const Tensor tensor, Args &&... args)
static std::unique_ptr< Operator >	Make (const Tensor tensor, Args &&... args)
static std::unique_ptr< Operator >	Make (const Config &config, const std::vector< Tensor > tensors, Args &&... args)
static std::unique_ptr< Operator >	Make (const std::vector< Tensor > tensors, Args &&... args)
static void	Call (const Handle &handle, const Config &config, const Args &... args)
static void	Call (const Tensor tensor, const Args &... args)
static auto	Call (const TensorLike &tensor, const Args &... args)
static void	Call (const Handle &handle, const Config &config, const Args &... args)
static void	Call (const Tensor tensor, const Args &... args)
static auto	Call (const TensorLike &tensor, const Args &... args)
static std::vector< std::size_t >	active_implementation_indices (Device::Type dev_type)
static std::vector< std::size_t >	active_implementation_indices (Device::Type dev_type)
Static Protected Attributes inherited from infini::ops::Operator< TopkSoftmax >
static constexpr Device::Type	device_type_
static constexpr std::size_t	implementation_index_

Constructor & Destructor Documentation

TopkSoftmax()

infini::ops::TopkSoftmax::TopkSoftmax ( const Tensor  gating_output, std::optional< Tensor >  bias, std::optional< Tensor >  is_padding, const bool  renormalize, Tensor  topk_weights, Tensor  topk_indices, Tensor  token_expert_indices  )

inline

Member Function Documentation

operator()()

virtual void infini::ops::TopkSoftmax::operator() ( const Tensor  gating_output, std::optional< Tensor >  bias, std::optional< Tensor >  is_padding, const bool  renormalize, Tensor  topk_weights, Tensor  topk_indices, Tensor  token_expert_indices  ) const

pure virtual

ValidateCallMetadata()

void infini::ops::TopkSoftmax::ValidateCallMetadata ( const Tensor  gating_output, std::optional< Tensor >  bias, std::optional< Tensor >  is_padding, const bool  renormalize, Tensor  topk_weights, Tensor  topk_indices, Tensor  token_expert_indices  ) const

inlineprotected

Member Data Documentation

device_index_

int infini::ops::TopkSoftmax::device_index_ {0}

protected

index_dtype_

DataType infini::ops::TopkSoftmax::index_dtype_

protected

input_dtype_

DataType infini::ops::TopkSoftmax::input_dtype_

protected

num_experts_

Tensor::Size infini::ops::TopkSoftmax::num_experts_ {0}

protected

num_tokens_

Tensor::Size infini::ops::TopkSoftmax::num_tokens_ {0}

protected

renormalize_

bool infini::ops::TopkSoftmax::renormalize_ {false}

protected

topk_

Tensor::Size infini::ops::TopkSoftmax::topk_ {0}

protected

---

The documentation for this class was generated from the following file:

• topk_softmax.h