Struct QuestionAnsweringConfig

pub struct QuestionAnsweringConfig {

    pub model_resource: ModelResource,
    pub config_resource: Box<dyn ResourceProvider + Send>,
    pub vocab_resource: Box<dyn ResourceProvider + Send>,
    pub merges_resource: Option<Box<dyn ResourceProvider + Send>>,
    pub device: Device,
    pub model_type: ModelType,
    pub lower_case: bool,
    pub strip_accents: Option<bool>,
    pub add_prefix_space: Option<bool>,
    pub max_seq_length: usize,
    pub doc_stride: usize,
    pub max_query_length: usize,
    pub max_answer_length: usize,
    pub kind: Option<Kind>,
}

Configuration for question answering

Contains information regarding the model to load and device to place the model on.

Fields

model_resource: ModelResource

Model weights resource (default: pretrained DistilBERT model on SQuAD)

config_resource: Box<dyn ResourceProvider + Send>

Config resource (default: pretrained DistilBERT model on SQuAD)

vocab_resource: Box<dyn ResourceProvider + Send>

Vocab resource (default: pretrained DistilBERT model on SQuAD)

merges_resource: Option<Box<dyn ResourceProvider + Send>>

Merges resource (default: None)

device: Device

Device to place the model on (default: CUDA/GPU when available)

model_type: ModelType

Model type

lower_case: bool

Flag indicating if the model expects a lower casing of the input

strip_accents: Option<bool>

Flag indicating if the tokenizer should strip accents (normalization). Only used for BERT / ALBERT models

add_prefix_space: Option<bool>

Flag indicating if the tokenizer should add a white space before each tokenized input (needed for some Roberta models)

max_seq_length: usize

Maximum sequence length for the combined query and context

doc_stride: usize

Stride to apply if the context needs to be broken down due to a large length. Represents the number of overlapping tokens between sliding windows.

max_query_length: usize

Maximum length for the query

max_answer_length: usize

Maximum length for the answer

kind: Option<Kind>

Model weights precision. If not provided, will default to full precision on CPU, or the loaded weights precision otherwise

Implementations

impl QuestionAnsweringConfig

pub fn new<RC, RV>( model_type: ModelType, model_resource: ModelResource, config_resource: RC, vocab_resource: RV, merges_resource: Option<RV>, lower_case: bool, strip_accents: impl Into<Option<bool>>, add_prefix_space: impl Into<Option<bool>>, ) -> QuestionAnsweringConfig

where RC: ResourceProvider + Send + 'static, RV: ResourceProvider + Send + 'static,

Instantiate a new question answering configuration of the supplied type.

Arguments

• model_type - ModelType indicating the model type to load (must match with the actual data to be loaded!)
• model_resource - The ResourceProvider pointing to the model to load (e.g. model.ot)
• config_resource - The ResourceProvider pointing to the model configuration to load (e.g. config.json)
• vocab_resource - The ResourceProvider pointing to the tokenizer’s vocabulary to load (e.g. vocab.txt/vocab.json)
• merges_resource - An optional ResourceProvider pointing to the tokenizer’s merge file to load (e.g. merges.txt), needed only for Roberta.
• lower_case - A bool indicating whether the tokenizer should lower case all input (in case of a lower-cased model)

pub fn custom_new<RC, RV>( model_type: ModelType, model_resource: ModelResource, config_resource: RC, vocab_resource: RV, merges_resource: Option<RV>, lower_case: bool, strip_accents: impl Into<Option<bool>>, add_prefix_space: impl Into<Option<bool>>, max_seq_length: impl Into<Option<usize>>, doc_stride: impl Into<Option<usize>>, max_query_length: impl Into<Option<usize>>, max_answer_length: impl Into<Option<usize>>, ) -> QuestionAnsweringConfig

where RC: ResourceProvider + Send + 'static, RV: ResourceProvider + Send + 'static,

Instantiate a new question answering configuration of the supplied type.

Arguments

• model_type - ModelType indicating the model type to load (must match with the actual data to be loaded!)
• model_resource - The ResourceProvider pointing to the model to load (e.g. model.ot)
• config_resource - The ResourceProvider pointing to the model configuration to load (e.g. config.json)
• vocab_resource - The ResourceProvider pointing to the tokenizer’s vocabulary to load (e.g. vocab.txt/vocab.json)
• merges_resource - An optional ResourceProvider pointing to the tokenizer’s merge file to load (e.g. merges.txt), needed only for Roberta.
• lower_case - A bool indicating whether the tokenizer should lower case all input (in case of a lower-cased model)
• max_seq_length - Optional maximum sequence token length to limit memory footprint. If the context is too long, it will be processed with sliding windows. Defaults to 384.
• max_query_length - Optional maximum question token length. Defaults to 64.
• doc_stride - Optional stride to apply if a sliding window is required to process the input context. Represents the number of overlapping tokens between sliding windows. This should be lower than the max_seq_length minus max_query_length (otherwise there is a risk for the sliding window not to progress). Defaults to 128.
• max_answer_length - Optional maximum token length for the extracted answer. Defaults to 15.

Trait Implementations

impl Default for QuestionAnsweringConfig

fn default() -> QuestionAnsweringConfig

Returns the “default value” for a type.