The number of pretrained Large Language Models (LLMs) is increasing steadily, though the majority are designed predominantly for the English language. While state-of-the-art LLMs can handle other languages, due to language contamination or some degree of multilingual pretraining data, they are not optimized for non-English languages, leading to inefficient encoding (high token “fertility”) and slower inference speed.In this work, we thoroughly compare a variety of vocabulary adaptation techniques for optimizing English LLMs for the Italian language, and put forward Semantic Alignment Vocabulary Adaptation (SAVA), a novel method that leverages neural mapping for vocabulary substitution. SAVA achieves competitive performance across multiple downstream tasks, enhancing grounded alignment strategies. We adapt two LLMs: Mistral-7B-v0.1, reducing token fertility by 25%, and Llama-3.1-8B, optimizing the vocabulary and reducing the number of parameters by 1 billion. We show that, following the adaptation of the vocabulary, these models can recover their performance with a relatively limited stage of continual training on the target language. Finally, we test the capabilities of the adapted models on various multi-choice and generative tasks.

One of the most widely used tasks for evaluating Large Language Models (LLMs) is Multiple-Choice Question Answering (MCQA). While open-ended question answering tasks are more challenging to evaluate, MCQA tasks are, in principle, easier to assess, as the model’s answer is thought to be simple to extract and is compared directly to a set of predefined choices. However, recent studies have started to question the reliability of MCQA evaluation, showing that multiple factors can significantly impact the reported performance of LLMs, especially when the model generates free-form text before selecting one of the answer choices. In this work, we shed light on the inconsistencies of MCQA evaluation strategies, which can lead to inaccurate and misleading model comparisons. We systematically analyze whether existing answer extraction methods are aligned with human judgment, and how they are influenced by answer constraints in the prompt across different domains. Our experiments demonstrate that traditional evaluation strategies often underestimate LLM capabilities, while LLM-based answer extractors are prone to systematic errors. Moreover, we reveal a fundamental trade-off between including format constraints in the prompt to simplify answer extraction and allowing models to generate free-form text to improve reasoning. Our findings call for standardized evaluation methodologies and highlight the need for more reliable and consistent MCQA evaluation practices.

Recent Large Language Models (LLMs) have shown impressive performance in addressing complex aspects of human language. These models have also demonstrated significant capabilities in processing and generating Italian text, achieving state-of-the-art results on current benchmarks for the Italian language. However, the number of such benchmarks is still insufficient. A case in point is the “Open Ita LLM Leaderboard” which only supports three benchmarks, despite being one of the most popular evaluation suite for the evaluation of Italian-speaking LLMs. In this paper, we analyze the current pitfalls of existing evaluation suites and propose two ways to this gap: i) a new suite of automatically-translated benchmarks, drawn from the most popular English benchmarks; and ii) the adaptation of existing manual dataset so that they can be used to complement the evaluation of Italian LLMs. We discuss the pros and cons of both approaches and release all our data to foster further research on the evaluation of Italian-speaking LLMs.

The increasing popularity of Large Language Models (LLMs) has led to a surge in research on adapting existing models to different languages. However, the pretraining of non-English LLMs is still an underexplored area and there is no open-source endeavor that explores what is achievable with open Italian data. To address this issue, we present Minerva, the first family of LLMs trained from scratch on Italian data. The creation of Minerva is an opportunity to explore and investigate the pretraining of LLMs for the Italian language, outlining the challenges that arise when training LLMs with native Italian texts. Minerva demonstrates that an LLM for a specific language brings a number of practical benefits compared to the adaptation of an existing one, including deep control over the composition of the vocabulary and the training data. With this paper, we aim to provide a comprehensive overview of the design choices, results, and evaluation of our Minerva models, showing promising results on Italian benchmarks and downstream tasks. Most importantly, we share what we learned and the findings obtained during the development of Minerva, as we believe that our experience will be valuable for the academic and industrial communities interested in training non-English LLMs from scratch.