SmolKalam: Ensemble Quality-Filtered Translation at Scale for High Quality Arabic Post-Training Data

TL;DR

This paper introduces SmolKalam, a high-quality Arabic dataset created through ensemble translation and filtering, addressing the need for large-scale, reasoning-capable Arabic data for post-training language models.

Contribution

It presents a novel multi-model ensemble translation pipeline with quality filtering to produce superior Arabic datasets for language model post-training.

Findings

Ensemble translation improves dataset quality.

Quality filtering enhances translation accuracy.

Effective translation techniques for decoder-only models are identified.

Abstract

Although the community has tackled the acquisition of high-quality Arabic pretraining data, we still lack large-scale, multi-turn Arabic datasets that include reasoning and tool calling. Naive translation can work at the pretraining scale, but post-training demands much higher quality, which requires a stricter approach to dataset curation. In this work, we introduce SmolKalam, a translation of Smoltalk2 that uses a multi-model ensemble translation pipeline, applies quality filtering, and examines effective translation techniques for traditional decoder-only models through ablations.