ChocoLlama: Lessons Learned From Teaching Llamas Dutch

TL;DR

This paper investigates methods for adapting English-centric LLMs, specifically Llama-2 and Llama-3, to Dutch by collecting Dutch data, applying continued pretraining with LoRA, and experimenting with tokenizers, revealing insights into effective language adaptation strategies.

Contribution

The study compares adaptation techniques for Llama models to Dutch, highlighting the effectiveness of LoRA and tokenizer modifications, and provides a new Dutch benchmark for LLM evaluation.

Findings

LoRA effectively scales for language adaptation

Tokenizer modification with reinitialization improves performance

Llama-3 outperforms adapted Llama-2 in Dutch capabilities

Abstract

While Large Language Models (LLMs) have shown remarkable capabilities in natural language understanding and generation, their performance often lags in lower-resource, non-English languages due to biases in the training data. In this work, we explore strategies for adapting the primarily English LLMs (Llama-2 and Llama-3) to Dutch, a language spoken by 30 million people worldwide yet often underrepresented in LLM development. We collect 104GB of Dutch text ($32$B tokens) from various sources to first apply continued pretraining using low-rank adaptation (LoRA), complemented with Dutch posttraining strategies provided by prior work. For Llama-2, we consider using (i) the tokenizer of the original model, and (ii) training a new, Dutch-specific tokenizer combined with embedding reinitialization. We evaluate our adapted models, ChocoLlama-2, both on standard benchmarks and a novel Dutch benchmark, ChocoLlama-Bench. Our results demonstrate that LoRA can effectively scale for language adaptation, and that tokenizer modification with careful weight reinitialization can improve performance. Notably, Llama-3 was released during the course of this project and, upon evaluation, demonstrated superior Dutch capabilities compared to our Dutch-adapted versions of Llama-2. We hence apply the same adaptation technique to Llama-3, using its original tokenizer. While our adaptation methods enhanced Llama-2's Dutch capabilities, we found limited gains when applying the same techniques to Llama-3. This suggests that for ever improving, multilingual foundation models, language adaptation techniques may benefit more from focusing on language-specific posttraining rather than on continued pretraining. We hope this work contributes to the broader understanding of adapting LLMs to lower-resource languages, and to the development of Dutch LLMs in particular.