wechsel

Wechsel

Código para WeChsel: inicialización efectiva de integración de subvenciones para la transferencia interlingüística de modelos de lenguaje monolingüe publicados en NAACL2022.

Documento: https://aclanthology.org/2022.naacl-main.293/

Los modelos del papel están disponibles en el centro de Huggingface:

• roberta-base-wechsel-french
• roberta-base-wechsel-german
• roberta-base-wechsel-chinese
• roberta-base-wechsel-swahili
• gpt2-wechsel-french
• gpt2-wechsel-german
• gpt2-wechsel-chinese
• gpt2-wechsel-swahili

Instalación

Distribuimos un paquete Python a través de Pypi:

 pip install wechsel

Alternativamente, clone el repositorio, instale requirements.txt y ejecute el código en wechsel/ .

Uso de ejemplo

Transferir inglés roberta-base a Swahili:

 import torch
from transformers import AutoModel , AutoTokenizer
from datasets import load_dataset
from wechsel import WECHSEL , load_embeddings

source_tokenizer = AutoTokenizer . from_pretrained ( "roberta-base" )
model = AutoModel . from_pretrained ( "roberta-base" )

target_tokenizer = source_tokenizer . train_new_from_iterator (
    load_dataset ( "oscar" , "unshuffled_deduplicated_sw" , split = "train" )[ "text" ],
    vocab_size = len ( source_tokenizer )
)

wechsel = WECHSEL (
    load_embeddings ( "en" ),
    load_embeddings ( "sw" ),
    bilingual_dictionary = "swahili"
)

target_embeddings , info = wechsel . apply (
    source_tokenizer ,
    target_tokenizer ,
    model . get_input_embeddings (). weight . detach (). numpy (),
)

model . get_input_embeddings (). weight . data = torch . from_numpy ( target_embeddings )
model . config . vocab_size = len ( target_embeddings )

# if the model has separate output embeddings, also copy those
if not model . config . tie_word_embeddings :
    target_out_embeddings , info = wechsel . apply (
        source_tokenizer ,
        target_tokenizer ,
        model . get_output_embeddings (). weight . detach (). numpy (),
    )
    model . get_output_embeddings (). weight . data = torch . from_numpy ( target_out_embeddings )

# use `model` and `target_tokenizer` to continue training in Swahili! 

Diccionarios bilingües

Distribuimos 3276 diccionarios bilingües del inglés a otros idiomas para usar con Wechsel en dicts/ .

Citación

Por favor cita a Wechsel como

 @inproceedings{minixhofer-etal-2022-wechsel,
    title = "{WECHSEL}: Effective initialization of subword embeddings for cross-lingual transfer of monolingual language models",
    author = "Minixhofer, Benjamin  and
      Paischer, Fabian  and
      Rekabsaz, Navid",
    booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jul,
    year = "2022",
    address = "Seattle, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.naacl-main.293",
    pages = "3992--4006",
    abstract = "Large pretrained language models (LMs) have become the central building block of many NLP applications. Training these models requires ever more computational resources and most of the existing models are trained on English text only. It is exceedingly expensive to train these models in other languages. To alleviate this problem, we introduce a novel method {--} called WECHSEL {--} to efficiently and effectively transfer pretrained LMs to new languages. WECHSEL can be applied to any model which uses subword-based tokenization and learns an embedding for each subword. The tokenizer of the source model (in English) is replaced with a tokenizer in the target language and token embeddings are initialized such that they are semantically similar to the English tokens by utilizing multilingual static word embeddings covering English and the target language. We use WECHSEL to transfer the English RoBERTa and GPT-2 models to four languages (French, German, Chinese and Swahili). We also study the benefits of our method on very low-resource languages. WECHSEL improves over proposed methods for cross-lingual parameter transfer and outperforms models of comparable size trained from scratch with up to 64x less training effort. Our method makes training large language models for new languages more accessible and less damaging to the environment. We make our code and models publicly available.",
}

Expresiones de gratitud

Investigación apoyada con TPUS de Cloud de la TPU Research Cloud (TRC) de Google. Agradecemos a Andy Koh y Artus Krohn-Grimberghe por proporcionar recursos computacionales adicionales. La Unidad Ellis Linz, el Laboratorio de IA Lit, el Instituto para el Aprendizaje Autor, es apoyada por el Estado Federal Alto Austria. Agradecemos al proyecto Incontrol-RL (FFG-881064).