---
base_model:
- meta-llama/Llama-3.2-1B-Instruct
tags:
- taiwan
- local_knowledge
- chinese
- traditional_chinese
- llama3.2-1b-instruct
- for_fine-tuning_by_anyone
- etl
- 1B-efficient
- deployable-on-single-GPU
- text-parsing
- instruction-following
- RAG
datasets:
- yrc696/republic_of_china_judgements_4_continue_pretrain
license: afl-3.0
---

# ETLCH 介紹
原始模型為llama3.2-1b-instruct
由國立清華大學博士班遲佑成(1), 陽明交通大學段明濤(2) 臺北市立大學侯詠皓(2) 業餘繼續預訓練與微調而成，供公眾研究擴展知識邊界用，亦可供後續微調使用。
本模型強化中文輸出穩定性。相對llama3.2-1b-instruct顯著提升，統計學上達顯著性(<.05))。
請註明作者與詳細來源。非常謝謝!

Paper 請參考: https://arxiv.org/abs/2510.01616



原始英翻中
![image](https://cdn-uploads.huggingface.co/production/uploads/661e4de160f6ec9834be7f09/CcnywuZLZ2JN5A14RxSZ5.png)

原始RAG
![image](https://cdn-uploads.huggingface.co/production/uploads/661e4de160f6ec9834be7f09/SUtjYOWUsk374TG1-mRrC.png)

fine-tuning-英翻中

![image](https://cdn-uploads.huggingface.co/production/uploads/661e4de160f6ec9834be7f09/HlY1aW9HJ0gEwSNPKThlI.png)


fine-tuning-RAG

![image](https://cdn-uploads.huggingface.co/production/uploads/661e4de160f6ec9834be7f09/DgmOkVhgJs0nzsdlz3YFk.png)