infini::ops::PagedCachingInfinilm Class Reference

#include <paged_caching_infinilm.h>

Inheritance diagram for infini::ops::PagedCachingInfinilm:

Public Member Functions
	PagedCachingInfinilm (const Tensor k, const Tensor v, const Tensor slot_mapping, Tensor k_cache, Tensor v_cache)
virtual void	operator() (const Tensor k, const Tensor v, const Tensor slot_mapping, Tensor k_cache, Tensor v_cache) const =0
Public Member Functions inherited from infini::ops::Operator< PagedCachingInfinilm >
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 Attributes
DataType	dtype_
std::size_t	num_tokens_ {0}
std::size_t	num_kv_heads_ {0}
std::size_t	head_size_ {0}
std::size_t	block_size_ {0}
Tensor::Stride	k_src_stride_ {0}
Tensor::Stride	v_src_stride_ {0}
Tensor::Stride	k_cache_block_stride_ {0}
Tensor::Stride	v_cache_block_stride_ {0}
Tensor::Stride	k_cache_head_stride_ {0}
Tensor::Stride	v_cache_head_stride_ {0}
Tensor::Stride	k_cache_slot_stride_ {0}
Tensor::Stride	v_cache_slot_stride_ {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< PagedCachingInfinilm >
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< PagedCachingInfinilm >
static constexpr Device::Type	device_type_
static constexpr std::size_t	implementation_index_

Detailed Description

Deprecated:

Migrate to an open-source-aligned operator when available. This interface will be removed in a future release.

Constructor & Destructor Documentation

PagedCachingInfinilm()

infini::ops::PagedCachingInfinilm::PagedCachingInfinilm ( const Tensor  k, const Tensor  v, const Tensor  slot_mapping, Tensor  k_cache, Tensor  v_cache  )

inline

Member Function Documentation

operator()()

virtual void infini::ops::PagedCachingInfinilm::operator() ( const Tensor  k, const Tensor  v, const Tensor  slot_mapping, Tensor  k_cache, Tensor  v_cache  ) const

pure virtual

Member Data Documentation

block_size_

std::size_t infini::ops::PagedCachingInfinilm::block_size_ {0}

protected

dtype_

DataType infini::ops::PagedCachingInfinilm::dtype_

protected

head_size_

std::size_t infini::ops::PagedCachingInfinilm::head_size_ {0}

protected

k_cache_block_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::k_cache_block_stride_ {0}

protected

k_cache_head_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::k_cache_head_stride_ {0}

protected

k_cache_slot_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::k_cache_slot_stride_ {0}

protected

k_src_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::k_src_stride_ {0}

protected

num_kv_heads_

std::size_t infini::ops::PagedCachingInfinilm::num_kv_heads_ {0}

protected

num_tokens_

std::size_t infini::ops::PagedCachingInfinilm::num_tokens_ {0}

protected

v_cache_block_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::v_cache_block_stride_ {0}

protected

v_cache_head_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::v_cache_head_stride_ {0}

protected

v_cache_slot_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::v_cache_slot_stride_ {0}

protected

v_src_stride_

Tensor::Stride infini::ops::PagedCachingInfinilm::v_src_stride_ {0}

protected

---

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

• paged_caching_infinilm.h