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

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Alrashed, Sultan, Helwe, Chadi, Orabona, Francesco
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908670872256512
author Alrashed, Sultan
Helwe, Chadi
Orabona, Francesco
author_facet Alrashed, Sultan
Helwe, Chadi
Orabona, Francesco
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18411
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SmolKalam: Ensemble Quality-Filtered Translation at Scale for High Quality Arabic Post-Training Data
Alrashed, Sultan
Helwe, Chadi
Orabona, Francesco
Computation and Language
Artificial Intelligence
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.
title SmolKalam: Ensemble Quality-Filtered Translation at Scale for High Quality Arabic Post-Training Data
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2511.18411