rust_bert/models/m2m_100/
mod.rs

1//! # M2M-100 (Fan et al.)
2//!
3//! Implementation of the M2M-100 language model ([Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) Fan, Bhosale, Schwenk, Ma, El-Kishky, Goyal, Baines, Celebi, Wenzel, Chaudhary, Goyal, Birch, Liptchinsky, Edunov, Grave, Auli, Joulin, 2020).
4//! The base model is implemented in the `m2m_100::M2M100Model` struct. The model also includes a language model head: `m2m_100::M2M100ForConditionalGeneration`
5//! implementing the common `generation_utils::LanguageGenerator` trait shared between the models used for generation (see `pipelines` for more information).
6//! This model allows for direct translation between 100 languages.
7//! The translation capabilities are illustrated in `examples/translation_m2m100`, run with `cargo run --example translation_m2m100`.
8//!
9//! # Model set-up and pre-trained weights loading
10//!
11//! All models expect the following resources:
12//! - Configuration file expected to have a structure following the [Transformers library](https://github.com/huggingface/transformers)
13//! - Model weights are expected to have a structure and parameter names following the [Transformers library](https://github.com/huggingface/transformers). A conversion using the Python utility scripts is required to convert the `.bin` weights to the `.ot` format.
14//! - `M2M100Tokenizer` using a `config.json` vocabulary and a `spiece.model` SentencePiece BPE model
15//!
16//! Pretrained models are available and can be downloaded using RemoteResources.
17//!
18//! ```no_run
19//! # fn main() -> anyhow::Result<()> {
20//! #
21//! use tch::{nn, Device};
22//! # use std::path::PathBuf;
23//! use rust_bert::m2m_100::{M2M100Config, M2M100Model};
24//! use rust_bert::resources::{LocalResource, ResourceProvider};
25//! use rust_bert::Config;
26//! use rust_tokenizers::tokenizer::M2M100Tokenizer;
27//!
28//! let config_resource = LocalResource {
29//!     local_path: PathBuf::from("path/to/config.json"),
30//! };
31//! let vocab_resource = LocalResource {
32//!     local_path: PathBuf::from("path/to/vocab.txt"),
33//! };
34//! let merges_resource = LocalResource {
35//!     local_path: PathBuf::from("path/to/spiece.model"),
36//! };
37//! let weights_resource = LocalResource {
38//!     local_path: PathBuf::from("path/to/model.ot"),
39//! };
40//! let config_path = config_resource.get_local_path()?;
41//! let vocab_path = vocab_resource.get_local_path()?;
42//! let merges_path = merges_resource.get_local_path()?;
43//! let weights_path = weights_resource.get_local_path()?;
44//!
45//! let device = Device::cuda_if_available();
46//! let mut vs = nn::VarStore::new(device);
47//! let tokenizer: M2M100Tokenizer = M2M100Tokenizer::from_files(
48//!     vocab_path.to_str().unwrap(),
49//!     merges_path.to_str().unwrap(),
50//!     false,
51//! )?;
52//! let config = M2M100Config::from_file(config_path);
53//! let m2m100_model = M2M100Model::new(&vs.root(), &config);
54//! vs.load(weights_path)?;
55//!
56//! # Ok(())
57//! # }
58//! ```
59
60mod attention;
61mod decoder;
62mod embeddings;
63mod encoder;
64mod m2m_100_model;
65
66pub use m2m_100_model::{
67    M2M100Config, M2M100ConfigResources, M2M100ForConditionalGeneration, M2M100Generator,
68    M2M100MergesResources, M2M100Model, M2M100ModelResources, M2M100SourceLanguages,
69    M2M100TargetLanguages, M2M100VocabResources,
70};
71
72pub use attention::LayerState;
rust_bert/models/m2m_100/mod.rs

rust_bert/models/m2m_100/
mod.rs
