Mila 0.13.48
Deep Neural Network Library
Loading...
Searching...
No Matches
Mila::Dnn::Serialization::PretrainedMetadata Struct Referenceexport

Metadata for pretrained model. More...

Public Attributes

std::string activation
std::string architecture
std::string attention_type
uint32_t embedding_dim
uint32_t hidden_dim
uint32_t max_seq_length
std::string model_name
float norm_epsilon
std::string norm_type
uint32_t num_heads
uint32_t num_kv_heads
uint32_t num_layers
std::string positional_encoding
float rope_theta
bool use_bias
uint32_t vocab_size

Detailed Description

Metadata for pretrained model.

Member Data Documentation

◆ activation

std::string Mila::Dnn::Serialization::PretrainedMetadata::activation

◆ architecture

std::string Mila::Dnn::Serialization::PretrainedMetadata::architecture

◆ attention_type

std::string Mila::Dnn::Serialization::PretrainedMetadata::attention_type

◆ embedding_dim

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::embedding_dim

◆ hidden_dim

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::hidden_dim

◆ max_seq_length

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::max_seq_length

◆ model_name

std::string Mila::Dnn::Serialization::PretrainedMetadata::model_name

◆ norm_epsilon

float Mila::Dnn::Serialization::PretrainedMetadata::norm_epsilon

◆ norm_type

std::string Mila::Dnn::Serialization::PretrainedMetadata::norm_type

◆ num_heads

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::num_heads

◆ num_kv_heads

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::num_kv_heads

◆ num_layers

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::num_layers

◆ positional_encoding

std::string Mila::Dnn::Serialization::PretrainedMetadata::positional_encoding

◆ rope_theta

float Mila::Dnn::Serialization::PretrainedMetadata::rope_theta

◆ use_bias

bool Mila::Dnn::Serialization::PretrainedMetadata::use_bias

◆ vocab_size

uint32_t Mila::Dnn::Serialization::PretrainedMetadata::vocab_size

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