Saved in:
Bibliographic Details
Main Authors: Csibi, Zsolt, Gortka, Bence György, Gyöngyössy, Natabara, Nagy, Kornél, Nemeskey, Dávid Márk, Sallai, Martin, Simonyi, András, Szekeres, András Márk, Palkó, Gábor
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.01244
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910020430462976
author Csibi, Zsolt
Gortka, Bence György
Gyöngyössy, Natabara
Nagy, Kornél
Nemeskey, Dávid Márk
Sallai, Martin
Simonyi, András
Szekeres, András Márk
Palkó, Gábor
author_facet Csibi, Zsolt
Gortka, Bence György
Gyöngyössy, Natabara
Nagy, Kornél
Nemeskey, Dávid Márk
Sallai, Martin
Simonyi, András
Szekeres, András Márk
Palkó, Gábor
contents We present Racka, a lightweight, continually pretrained large language model designed to bridge the resource gap between Hungarian and high-resource languages such as English and German. Racka employs parameter-efficient continual pretraining via Low-Rank Adaptation (LoRA) on a Qwen-3 4B backbone, making the recipe practical on A100 (40GB)-based HPC clusters with low inter-node bandwidth. To better match the training distribution, we replace and adapt the tokenizer, achieving substantially improved tokenization fertility for Hungarian while maintaining competitive performance in English and German. The model is trained on 160B subword tokens drawn from a mixture of internet and high-quality curated sources, with a composition of 44% Hungarian, 24% English, 21% German, and 11% code. This data mix is chosen to mitigate catastrophic forgetting and preserve high-resource language capabilities during continual pretraining. Our preliminary results indicate modest but stable results in language adaptation.
format Preprint
id arxiv_https___arxiv_org_abs_2601_01244
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Racka: Efficient Hungarian LLM Adaptation on Academic Infrastructure
Csibi, Zsolt
Gortka, Bence György
Gyöngyössy, Natabara
Nagy, Kornél
Nemeskey, Dávid Márk
Sallai, Martin
Simonyi, András
Szekeres, András Márk
Palkó, Gábor
Computation and Language
We present Racka, a lightweight, continually pretrained large language model designed to bridge the resource gap between Hungarian and high-resource languages such as English and German. Racka employs parameter-efficient continual pretraining via Low-Rank Adaptation (LoRA) on a Qwen-3 4B backbone, making the recipe practical on A100 (40GB)-based HPC clusters with low inter-node bandwidth. To better match the training distribution, we replace and adapt the tokenizer, achieving substantially improved tokenization fertility for Hungarian while maintaining competitive performance in English and German. The model is trained on 160B subword tokens drawn from a mixture of internet and high-quality curated sources, with a composition of 44% Hungarian, 24% English, 21% German, and 11% code. This data mix is chosen to mitigate catastrophic forgetting and preserve high-resource language capabilities during continual pretraining. Our preliminary results indicate modest but stable results in language adaptation.
title Racka: Efficient Hungarian LLM Adaptation on Academic Infrastructure
topic Computation and Language
url https://arxiv.org/abs/2601.01244