Struct EmbeddingBagFromPretrainedOptions¶
Defined in File embedding.h
Page Contents
Struct Documentation¶
-
struct EmbeddingBagFromPretrainedOptions¶
Options for the
EmbeddingBag::from_pretrained
function.Public Functions
-
inline auto freeze(const bool &new_freeze) -> decltype(*this)¶
If
true
, the tensor does not get updated in the learning process.Equivalent to
embeddingbag.weight.requires_grad_(false)
. Default:true
-
inline auto freeze(bool &&new_freeze) -> decltype(*this)¶
-
inline const bool &freeze() const noexcept¶
-
inline bool &freeze() noexcept¶
-
inline auto max_norm(const std::optional<double> &new_max_norm) -> decltype(*this)¶
If given, each embedding vector with norm larger than
max_norm
is renormalized to have normmax_norm
.
-
inline auto max_norm(std::optional<double> &&new_max_norm) -> decltype(*this)¶
-
inline const std::optional<double> &max_norm() const noexcept¶
-
inline std::optional<double> &max_norm() noexcept¶
-
inline auto norm_type(const double &new_norm_type) -> decltype(*this)¶
The p of the p-norm to compute for the
max_norm
option. Default2
.
-
inline auto norm_type(double &&new_norm_type) -> decltype(*this)¶
-
inline const double &norm_type() const noexcept¶
-
inline double &norm_type() noexcept¶
-
inline auto scale_grad_by_freq(const bool &new_scale_grad_by_freq) -> decltype(*this)¶
If given, this will scale gradients by the inverse of frequency of the words in the mini-batch.
Default
false
. Note: this option is not supported whenmode="kMax"
.
-
inline auto scale_grad_by_freq(bool &&new_scale_grad_by_freq) -> decltype(*this)¶
-
inline const bool &scale_grad_by_freq() const noexcept¶
-
inline bool &scale_grad_by_freq() noexcept¶
-
inline auto mode(const EmbeddingBagMode &new_mode) -> decltype(*this)¶
"kSum"
,"kMean"
or"kMax"
.Specifies the way to reduce the bag.
"kSum"
computes the weighted sum, takingper_sample_weights
into consideration."kMean"
computes the average of the values in the bag,"kMax"
computes the max value over each bag.
-
inline auto mode(EmbeddingBagMode &&new_mode) -> decltype(*this)¶
-
inline const EmbeddingBagMode &mode() const noexcept¶
-
inline EmbeddingBagMode &mode() noexcept¶
-
inline auto sparse(const bool &new_sparse) -> decltype(*this)¶
If
true
, gradient w.r.t.weight
matrix will be a sparse tensor. Note: this option is not supported whenmode="kMax"
.
-
inline auto sparse(bool &&new_sparse) -> decltype(*this)¶
-
inline const bool &sparse() const noexcept¶
-
inline bool &sparse() noexcept¶
-
inline auto include_last_offset(const bool &new_include_last_offset) -> decltype(*this)¶
If
true
,offsets
has one additional element, where the last element is equivalent to the size ofindices
.This matches the CSR format. Note: this option is currently only supported when
mode="sum"
.
-
inline auto include_last_offset(bool &&new_include_last_offset) -> decltype(*this)¶
-
inline const bool &include_last_offset() const noexcept¶
-
inline bool &include_last_offset() noexcept¶
-
inline auto padding_idx(const std::optional<int64_t> &new_padding_idx) -> decltype(*this)¶
If specified, the entries at
padding_idx
do not contribute to the gradient; therefore, the embedding vector at padding_idx is not updated during training, i.e.it remains as a fixed “pad”. Note that the embedding vector at
padding_idx
is excluded from the reduction.
-
inline auto padding_idx(std::optional<int64_t> &&new_padding_idx) -> decltype(*this)¶
-
inline const std::optional<int64_t> &padding_idx() const noexcept¶
-
inline std::optional<int64_t> &padding_idx() noexcept¶
-
inline auto freeze(const bool &new_freeze) -> decltype(*this)¶