Module rust_bert::t5

T5 (Text-To-Text Transfer Transformer)

Implementation of the T5 language model (Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer Raffel, Shazeer, Roberts, Lee, Narang, Matena, Zhou, Li, Liu, 2019). The base model is implemented in the t5::T5Model struct. This model includes a language model head: t5::T5ForConditionalGeneration implementing the common generation::LMHeadModel trait shared between the models used for generation (see pipelines for more information).

Model set-up and pre-trained weights loading

A full working example (translation) is provided in examples/t5.rs, run with cargo run --example t5. All models expect the following resources:

• Configuration file expected to have a structure following the Transformers library
• Model weights are expected to have a structure and parameter names following the Transformers library. A conversion using the Python utility scripts is required to convert the .bin weights to the .ot format.
• T5Tokenizer using a spiece.model sentence piece model

Pretrained models for a number of language pairs are available and can be downloaded using RemoteResources.

use tch::{nn, Device};
use rust_bert::resources::{download_resource, LocalResource, Resource};
use rust_bert::t5::{T5Config, T5ForConditionalGeneration};
use rust_bert::Config;
use rust_tokenizers::preprocessing::tokenizer::t5_tokenizer::T5Tokenizer;

let config_resource = Resource::Local(LocalResource {
    local_path: PathBuf::from("path/to/config.json"),
});
let sentence_piece_resource = Resource::Local(LocalResource {
    local_path: PathBuf::from("path/to/spiece.model"),
});
let weights_resource = Resource::Local(LocalResource {
    local_path: PathBuf::from("path/to/model.ot"),
});
let config_path = download_resource(&config_resource)?;
let spiece_path = download_resource(&sentence_piece_resource)?;
let weights_path = download_resource(&weights_resource)?;

let device = Device::cuda_if_available();
let mut vs = nn::VarStore::new(device);
let tokenizer = T5Tokenizer::from_file(spiece_path.to_str().unwrap(), true);
let config = T5Config::from_file(config_path);
let t5_model = T5ForConditionalGeneration::new(&vs.root(), &config, false, false);
vs.load(weights_path)?;

Structs

LayerState	Cache for T5 attention layers
T5Config	T5 model configuration
T5ConfigResources	T5 Pretrained model config files
T5ForConditionalGeneration	T5 Model for conditional generation
T5Model	T5 Base model
T5ModelResources	T5 Pretrained model weight files
T5Prefix	T5 optional prefixes
T5VocabResources	T5 Pretrained model vocab files