Evaluating Large Language Models with Human Feedback: Establishing a Swedish Benchmark

TL;DR

This paper introduces a Swedish language benchmark using human feedback to evaluate and compare the performance of various large language models, addressing a gap in resources for less-represented languages.

Contribution

It presents a new Swedish benchmark with human feedback for LLM evaluation and releases a tool to facilitate future research and model assessment in Swedish.

Findings

Evaluated 11 large language models on Swedish language tasks.

Provided a publicly available benchmark tool for Swedish LLM assessment.

Aims to establish a leaderboard for Swedish language models.

Abstract

In the rapidly evolving field of artificial intelligence, large language models (LLMs) have demonstrated significant capabilities across numerous applications. However, the performance of these models in languages with fewer resources, such as Swedish, remains under-explored. This study introduces a comprehensive human benchmark to assess the efficacy of prominent LLMs in understanding and generating Swedish language texts using forced choice ranking. We employ a modified version of the ChatbotArena benchmark, incorporating human feedback to evaluate eleven different models, including GPT-4, GPT-3.5, various Claude and Llama models, and bespoke models like Dolphin-2.9-llama3b-8b-flashback and BeagleCatMunin. These models were chosen based on their performance on LMSYS chatbot arena and the Scandeval benchmarks. We release the chatbotarena.se benchmark as a tool to improve our understanding of language model performance in Swedish with the hopes that it will be widely used. We aim to create a leaderboard once sufficient data has been collected and analysed.