# LianjiaTech/BELLE

BELLE: Be Everyone's Large Language model Engine（开源中文对话大模型）

Repository: https://github.com/LianjiaTech/BELLE
Canonical: https://ross.abutalabs.com/products/belle
Language: HTML
License: Apache-2.0
License Family: permissive
Topics: bloom, instruction-set, llama, open-models, gpt-q, instruct-gpt, gpt-evaluation, chinese-nlp, lora, instruct-finetune
Last push: 2024-10-16T11:38:59+00:00

## Health v2 (maintenance only)
Score: 21/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 90
- inputs: {"age_days": 1265, "days_push": 686, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 8275, forks 756 (observed 2026-08-28T04:10:20.024621+00:00)

## What it is
BELLE is an open-source project providing Chinese-optimized instruction-tuned large language models, training data, and fine-tuning code built on open pretrained backbones like LLaMA and BLOOM. It also releases enhanced Chinese speech recognition models (fine-tuned Whisper) and multimodal variants, along with technical reports on training techniques.

## Use cases
- fine-tune a Chinese instruction-following chatbot on an open LLM
- download Chinese instruction tuning datasets generated from ChatGPT
- train models with LoRA or RLHF (PPO/DPO) for Chinese dialogue
- improve Whisper speech recognition accuracy for Chinese audio
- explore multimodal Chinese vision-language models
- reduce the barrier to building custom Chinese LLMs

## When to choose
- you need Chinese-language instruction-tuned models or training data
- you want open recipes for LoRA, full fine-tuning, or RLHF on Chinese LLMs
- you need Chinese-optimized speech recognition models
- you are researching how training data quality affects LLM performance

## When to avoid
- you need a production-ready hosted chat service rather than models and training code
- your focus is English-only or multilingual models beyond Chinese optimization
- you need actively cutting-edge releases, as updates have slowed since late 2024

## Facets
- artifact type: learning-resource
- maturity: maintenance
- function: llm-training, machine-learning, speech-recognition, rag
- domain: large-language-models, machine-learning, speech-processing
- platform: python
- tags: chinese-nlp, instruction-tuning, lora, rlhf, whisper, multimodal, open-models, chatglm, llama, natural-language-processing, gpu, linux

## Member repositories
- LianjiaTech/BELLE (main) score 21

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:20.024621+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:28:07.026401+00:00, confidence not recorded.
  - readme: https://github.com/LianjiaTech/BELLE (fetched 2026-08-28T04:10:20.024621+00:00, sha 6c988e713502)
- Data as of 2026-08-30T08:39:29.467469+00:00.