Salvato in:
Dettagli Bibliografici
Autori principali: Chih, Yu-Cheng, Duan, Ming-Tao, Hou, Yong-Hao
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2510.01616
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916985341739008
author Chih, Yu-Cheng
Duan, Ming-Tao
Hou, Yong-Hao
author_facet Chih, Yu-Cheng
Duan, Ming-Tao
Hou, Yong-Hao
contents Small Language Models (SLMs) enable cost-effective, on-device and latency-sensitive AI applications, yet their deployment in Traditional Chinese (TC) remains hindered by token-level instability - models unpredictably emit non-TC characters or code-switch into other languages. We address this practical reliability gap by creating PureTC-1B, a three-stage stabilization pipeline for Llama-3.2-1B-Instruct (an open-weight, instruction-tuned model released by Meta) using parameter-efficient LoRA adapters. Our method combines Continual Pre-Training (CPT) on TC-centric corpora, Supervised Fine-Tuning (SFT) with instruction data, and Direct Preference Optimization (DPO) using TC-adherence preferences to improve monolingual robustness without full-model retraining. On a benchmark designed to simulate real-world usage, PureTC-1B achieves a 51.3% relative reduction (micro-average) in non-TC output tokens versus the base model. On a Named Entity Translation (NET) task, PureTC-1B further reduces incorrect-language tokens by 77.2% relative to Llama-3B and 57.2% relative to Qwen-1.5B, indicating that robust TC adherence is attainable even at the 1B scale. The pipeline is reproducible, adapter-only, and hardware-friendly, offering practitioners a practical recipe to enhance language stability for TC and potentially other non-English languages.
format Preprint
id arxiv_https___arxiv_org_abs_2510_01616
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Efficient Training of Robust Traditional Chinese LLaMA-1B on a Single Consumer GPU: Continual Pre-training, SFT, and DPO
Chih, Yu-Cheng
Duan, Ming-Tao
Hou, Yong-Hao
Computation and Language
Small Language Models (SLMs) enable cost-effective, on-device and latency-sensitive AI applications, yet their deployment in Traditional Chinese (TC) remains hindered by token-level instability - models unpredictably emit non-TC characters or code-switch into other languages. We address this practical reliability gap by creating PureTC-1B, a three-stage stabilization pipeline for Llama-3.2-1B-Instruct (an open-weight, instruction-tuned model released by Meta) using parameter-efficient LoRA adapters. Our method combines Continual Pre-Training (CPT) on TC-centric corpora, Supervised Fine-Tuning (SFT) with instruction data, and Direct Preference Optimization (DPO) using TC-adherence preferences to improve monolingual robustness without full-model retraining. On a benchmark designed to simulate real-world usage, PureTC-1B achieves a 51.3% relative reduction (micro-average) in non-TC output tokens versus the base model. On a Named Entity Translation (NET) task, PureTC-1B further reduces incorrect-language tokens by 77.2% relative to Llama-3B and 57.2% relative to Qwen-1.5B, indicating that robust TC adherence is attainable even at the 1B scale. The pipeline is reproducible, adapter-only, and hardware-friendly, offering practitioners a practical recipe to enhance language stability for TC and potentially other non-English languages.
title Efficient Training of Robust Traditional Chinese LLaMA-1B on a Single Consumer GPU: Continual Pre-training, SFT, and DPO
topic Computation and Language
url https://arxiv.org/abs/2510.01616