infini::ops::GroupedTopk Class Reference

#include <grouped_topk.h>

Inheritance diagram for infini::ops::GroupedTopk:

Public Member Functions
	GroupedTopk (const Tensor scores, const Tensor bias, const int64_t num_expert_group, const int64_t topk_group, const int64_t topk, const bool renormalize, const double routed_scaling_factor, const int64_t scoring_func, Tensor topk_values, Tensor topk_indices)
virtual void	operator() (const Tensor scores, const Tensor bias, const int64_t num_expert_group, const int64_t topk_group, const int64_t topk, const bool renormalize, const double routed_scaling_factor, const int64_t scoring_func, Tensor topk_values, Tensor topk_indices) const =0
Public Member Functions inherited from infini::ops::Operator< GroupedTopk >
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 scores, const Tensor bias, const int64_t num_expert_group, const int64_t topk_group, const int64_t topk, const bool renormalize, const double routed_scaling_factor, const int64_t scoring_func, Tensor topk_values, Tensor topk_indices) const

Protected Attributes
Tensor::Size	num_tokens_ {0}
Tensor::Size	num_experts_ {0}
int64_t	num_expert_group_ {0}
int64_t	topk_group_ {0}
int64_t	topk_ {0}
bool	renormalize_ {false}
double	routed_scaling_factor_ {1.0}
int64_t	scoring_func_ {0}
DataType	scores_dtype_
DataType	bias_dtype_
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< GroupedTopk >
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< GroupedTopk >
static constexpr Device::Type	device_type_
static constexpr std::size_t	implementation_index_

Constructor & Destructor Documentation

GroupedTopk()

infini::ops::GroupedTopk::GroupedTopk ( const Tensor  scores, const Tensor  bias, const int64_t  num_expert_group, const int64_t  topk_group, const int64_t  topk, const bool  renormalize, const double  routed_scaling_factor, const int64_t  scoring_func, Tensor  topk_values, Tensor  topk_indices  )

inline

Member Function Documentation

operator()()

virtual void infini::ops::GroupedTopk::operator() ( const Tensor  scores, const Tensor  bias, const int64_t  num_expert_group, const int64_t  topk_group, const int64_t  topk, const bool  renormalize, const double  routed_scaling_factor, const int64_t  scoring_func, Tensor  topk_values, Tensor  topk_indices  ) const

pure virtual

ValidateCallMetadata()

void infini::ops::GroupedTopk::ValidateCallMetadata ( const Tensor  scores, const Tensor  bias, const int64_t  num_expert_group, const int64_t  topk_group, const int64_t  topk, const bool  renormalize, const double  routed_scaling_factor, const int64_t  scoring_func, Tensor  topk_values, Tensor  topk_indices  ) const

inlineprotected

Member Data Documentation

bias_dtype_

DataType infini::ops::GroupedTopk::bias_dtype_

protected

device_index_

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

protected

num_expert_group_

int64_t infini::ops::GroupedTopk::num_expert_group_ {0}

protected

num_experts_

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

protected

num_tokens_

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

protected

renormalize_

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

protected

routed_scaling_factor_

double infini::ops::GroupedTopk::routed_scaling_factor_ {1.0}

protected

scores_dtype_

DataType infini::ops::GroupedTopk::scores_dtype_

protected

scoring_func_

int64_t infini::ops::GroupedTopk::scoring_func_ {0}

protected

topk_

int64_t infini::ops::GroupedTopk::topk_ {0}

protected

topk_group_

int64_t infini::ops::GroupedTopk::topk_group_ {0}

protected

---

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

• grouped_topk.h