Advancing Large Language Models for Tibetan with Curated Data and Continual Pre-Training

• URL: http://arxiv.org/abs/2507.09205v4
• Date: Mon, 28 Jul 2025 04:56:37 GMT
• Title: Advancing Large Language Models for Tibetan with Curated Data and Continual Pre-Training
• Authors: Leiyu Pan, Bojian Xiong, Lei Yang, Renren Jin, Shaowei Zhang, Yue Chen, Ling Shi, Jiang Zhou, Junru Wu, Zhen Wang, Jianxiang Peng, Juesi Xiao, Tianyu Dong, Zhuowen Han, Zhuo Chen, Yuqi Ren, Deyi Xiong,
• Abstract summary: We create the largest Tibetan pre-training corpus to date, aggregating data from diverse sources.<n>With curated data, we continue pre/post-training a multilingual base model to enhance its generative capabilities in Tibetan.<n>We create new high-quality Tibetan benchmarks, and complement them with existing public benchmarks.
• Score: 43.57169338795754
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models have achieved remarkable progress across many languages. However, Tibetan, as a representative low-resource language, is particularly underrepresented in existing models due to the scarcity of high-quality training corpora. To address this gap, we curate the largest Tibetan pre-training corpus to date, aggregating data from diverse sources and applying a dedicated data cleaning and processing pipeline tailored for Tibetan. With the curated data, we continue pre/post-training a multilingual base model to enhance its generative capabilities in Tibetan. To evaluate the Tibetan capabilities of the model, we create new high-quality Tibetan benchmarks, and complement them with existing public benchmarks. Experimental results demonstrate that our model consistently and significantly outperforms both open-source models of similar scale and Tibetan-tailored models across a wide range of tasks.

Related papers

• TIB-STC: A Large-Scale Structured Tibetan Benchmark for Low-Resource Language Modeling [10.77750944881769]
  We present TIB-STC, the first large-scale, expert-curated, and multi-domain dataset specifically designed to support the development and evaluation of large language models (LLMs)<n>Spanning over 11 billion tokens across literature, religion, medicine, law, and daily communication, TIB-STC preserves traditional grammar and stylistic richness.<n> Evaluation on TLUE Benchmark for Tibetan-specific tasks, including Ti-MMLU and Ti-SafetyBench, demonstrates the TIB-STC's effectiveness in enabling robust instruction-following and culturally aligned generation.
  arXiv  Detail & Related papers  (2025-03-24T02:17:41Z)
• PEFTT: Parameter-Efficient Fine-Tuning for low-resource Tibetan pre-trained language models [0.0]
  There is currently no existing large language model for Tibetan due to its low-resource nature. We conducted three types of efficient fine-tuning experiments on the publicly available TNCC-title dataset.
  arXiv  Detail & Related papers  (2023-09-21T14:29:23Z)
• WanJuan: A Comprehensive Multimodal Dataset for Advancing English and Chinese Large Models [69.96148259273065]
  "Wan Juan" is a large-scale multimodal dataset composed of both Chinese and English data, collected from a wide range of web sources. It was utilized in the training of InternLM, a model that demonstrated significant advantages in multi-dimensional evaluations when compared to models of a similar scale.
  arXiv  Detail & Related papers  (2023-08-21T14:40:48Z)
• Fine-grained Text Style Transfer with Diffusion-Based Language Models [50.02698074338317]
  We trained a diffusion-based model on StylePTB dataset, the standard benchmark for fine-grained text style transfers. Our model was able to achieve state-of-the-art performance on both individual and compositional transfers.
  arXiv  Detail & Related papers  (2023-05-31T02:51:26Z)
• Pre-training Data Quality and Quantity for a Low-Resource Language: New Corpus and BERT Models for Maltese [4.4681678689625715]
  We analyse the effect of pre-training with monolingual data for a low-resource language. We present a newly created corpus for Maltese, and determine the effect that the pre-training data size and domain have on the downstream performance. We compare two models on the new corpus: a monolingual BERT model trained from scratch (BERTu), and a further pre-trained multilingual BERT (mBERTu)
  arXiv  Detail & Related papers  (2022-05-21T06:44:59Z)
• TiBERT: Tibetan Pre-trained Language Model [2.9554549423413303]
  This paper collects the large-scale training data from Tibetan websites and constructs a vocabulary that can cover 99.95$%$ of the words in the corpus by using Sentencepiece. We apply TiBERT to the downstream tasks of text classification and question generation, and compare it with classic models and multilingual pre-trained models.
  arXiv  Detail & Related papers  (2022-05-15T14:45:08Z)
• Can Character-based Language Models Improve Downstream Task Performance in Low-Resource and Noisy Language Scenarios? [15.995677143912474]
  We focus on North-African colloquial dialectal Arabic written using an extension of the Latin script, called NArabizi.<n>We show that a character-based model trained on only 99k sentences of NArabizi and fined-tuned on a small treebank leads to performance close to those obtained with the same architecture pre-trained on large multilingual and monolingual models.
  arXiv  Detail & Related papers  (2021-10-26T14:59:16Z)
• bert2BERT: Towards Reusable Pretrained Language Models [51.078081486422896]
  We propose bert2BERT, which can effectively transfer the knowledge of an existing smaller pre-trained model to a large model. bert2BERT saves about 45% and 47% computational cost of pre-training BERT_BASE and GPT_BASE by reusing the models of almost their half sizes.
  arXiv  Detail & Related papers  (2021-10-14T04:05:25Z)
• Unsupervised Domain Adaptation of a Pretrained Cross-Lingual Language Model [58.27176041092891]
  Recent research indicates that pretraining cross-lingual language models on large-scale unlabeled texts yields significant performance improvements. We propose a novel unsupervised feature decomposition method that can automatically extract domain-specific features from the entangled pretrained cross-lingual representations. Our proposed model leverages mutual information estimation to decompose the representations computed by a cross-lingual model into domain-invariant and domain-specific parts.
  arXiv  Detail & Related papers  (2020-11-23T16:00:42Z)
• Unsupervised Paraphrasing with Pretrained Language Models [85.03373221588707]
  We propose a training pipeline that enables pre-trained language models to generate high-quality paraphrases in an unsupervised setting. Our recipe consists of task-adaptation, self-supervision, and a novel decoding algorithm named Dynamic Blocking. We show with automatic and human evaluations that our approach achieves state-of-the-art performance on both the Quora Question Pair and the ParaNMT datasets.
  arXiv  Detail & Related papers  (2020-10-24T11:55:28Z)
• Harnessing Multilinguality in Unsupervised Machine Translation for Rare Languages [48.28540903568198]
  We show that multilinguality is critical to making unsupervised systems practical for low-resource settings. We present a single model for 5 low-resource languages (Gujarati, Kazakh, Nepali, Sinhala, and Turkish) to and from English directions. We outperform all current state-of-the-art unsupervised baselines for these languages, achieving gains of up to 14.4 BLEU.
  arXiv  Detail & Related papers  (2020-09-23T15:07:33Z)
• ParsBERT: Transformer-based Model for Persian Language Understanding [0.7646713951724012]
  This paper proposes a monolingual BERT for the Persian language (ParsBERT) It shows its state-of-the-art performance compared to other architectures and multilingual models. ParsBERT obtains higher scores in all datasets, including existing ones as well as composed ones.
  arXiv  Detail & Related papers  (2020-05-26T05:05:32Z)
• Exploring Fine-tuning Techniques for Pre-trained Cross-lingual Models via Continual Learning [74.25168207651376]
  Fine-tuning pre-trained language models to downstream cross-lingual tasks has shown promising results. We leverage continual learning to preserve the cross-lingual ability of the pre-trained model when we fine-tune it to downstream tasks. Our methods achieve better performance than other fine-tuning baselines on the zero-shot cross-lingual part-of-speech tagging and named entity recognition tasks.
  arXiv  Detail & Related papers  (2020-04-29T14:07:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.