BELLE

BELLE (Be Everyone’s Large Language model Engine) is an open-source Chinese large language model project from LianjiaTech, published on GitHub with around 8,300 stars. Launched in the early wave of open Chinese LLMs, it combined instruction-tuning data, training code, and released models to help the community build capable Chinese conversational models.

Overview

BELLE’s central contribution was data. The team generated millions of Chinese instruction examples using ChatGPT, following the self-instruct approach, and released them openly so others could fine-tune base models such as BLOOMZ and LLaMA into Chinese chat assistants. Alongside the datasets, the project published fine-tuned model weights at several sizes, the fine-tuning code (including LoRA-based recipes), and evaluation work comparing model outputs. At a time when high-quality Chinese instruction data was scarce, BELLE’s releases were widely used by Chinese open-source projects and researchers to bootstrap their own models. The repository’s main activity ran through 2024, and it remains a well-documented reference for Chinese instruction tuning: data generation pipelines, training scripts, model cards, and lessons learned are all preserved in the open.

Key Features

• Millions of openly released Chinese instruction-tuning examples
• Fine-tuned Chinese chat models based on BLOOMZ and LLaMA
• Complete training code including LoRA fine-tuning recipes
• Evaluation benchmarks and analysis for Chinese model quality
• Data generation pipeline based on the self-instruct method
• Apache-2.0 licensed and widely reused by the Chinese LLM community

💳 Pricing

BELLE is free and open source under the Apache-2.0 license. The datasets, training code, and model weights are all available from the GitHub repository at no cost.

Who Is It For

BELLE is for researchers and developers working on Chinese language models — especially anyone who needs open Chinese instruction data or a proven, documented recipe for fine-tuning base models into Chinese assistants.