Guardado en:
Detalles Bibliográficos
Autores principales: Mundra, Nandini, Kishore, Aditya Nanda, Dabre, Raj, Puduppully, Ratish, Kunchukuttan, Anoop, Khapra, Mitesh M.
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:https://arxiv.org/abs/2407.05841
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913558805086208
author Mundra, Nandini
Kishore, Aditya Nanda
Dabre, Raj
Puduppully, Ratish
Kunchukuttan, Anoop
Khapra, Mitesh M.
author_facet Mundra, Nandini
Kishore, Aditya Nanda
Dabre, Raj
Puduppully, Ratish
Kunchukuttan, Anoop
Khapra, Mitesh M.
contents Language Models (LMs) excel in natural language processing tasks for English but show reduced performance in most other languages. This problem is commonly tackled by continually pre-training and fine-tuning these models for said languages. A significant issue in this process is the limited vocabulary coverage in the original model's tokenizer, leading to inadequate representation of new languages and necessitating an expansion of the tokenizer. The initialization of the embeddings corresponding to new vocabulary items presents a further challenge. Current strategies require cross-lingual embeddings and lack a solid theoretical foundation as well as comparisons with strong baselines. In this paper, we first establish theoretically that initializing within the convex hull of existing embeddings is a good initialization, followed by a novel but simple approach, Constrained Word2Vec (CW2V), which does not require cross-lingual embeddings. Our study evaluates different initialization methods for expanding RoBERTa and LLaMA 2 across four languages and five tasks. The results show that CW2V performs equally well or even better than more advanced techniques. Additionally, simpler approaches like multivariate initialization perform on par with these advanced methods indicating that efficient large-scale multilingual continued pretraining can be achieved even with simpler initialization methods. We release our code publicly (https://github.com/AI4Bharat/VocabAdaptation_LLM/tree/CW2V).
format Preprint
id arxiv_https___arxiv_org_abs_2407_05841
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle An Empirical Comparison of Vocabulary Expansion and Initialization Approaches for Language Models
Mundra, Nandini
Kishore, Aditya Nanda
Dabre, Raj
Puduppully, Ratish
Kunchukuttan, Anoop
Khapra, Mitesh M.
Computation and Language
Machine Learning
Language Models (LMs) excel in natural language processing tasks for English but show reduced performance in most other languages. This problem is commonly tackled by continually pre-training and fine-tuning these models for said languages. A significant issue in this process is the limited vocabulary coverage in the original model's tokenizer, leading to inadequate representation of new languages and necessitating an expansion of the tokenizer. The initialization of the embeddings corresponding to new vocabulary items presents a further challenge. Current strategies require cross-lingual embeddings and lack a solid theoretical foundation as well as comparisons with strong baselines. In this paper, we first establish theoretically that initializing within the convex hull of existing embeddings is a good initialization, followed by a novel but simple approach, Constrained Word2Vec (CW2V), which does not require cross-lingual embeddings. Our study evaluates different initialization methods for expanding RoBERTa and LLaMA 2 across four languages and five tasks. The results show that CW2V performs equally well or even better than more advanced techniques. Additionally, simpler approaches like multivariate initialization perform on par with these advanced methods indicating that efficient large-scale multilingual continued pretraining can be achieved even with simpler initialization methods. We release our code publicly (https://github.com/AI4Bharat/VocabAdaptation_LLM/tree/CW2V).
title An Empirical Comparison of Vocabulary Expansion and Initialization Approaches for Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2407.05841