{"data":{"slug":"lianjiatech-belle","name":"BELLE","tagline":"Open-source Chinese dialogue large model","github_url":"https://github.com/LianjiaTech/BELLE","owner":"LianjiaTech","repo":"BELLE","owner_avatar_url":"https://avatars.githubusercontent.com/u/14540911?v=4","primary_language":"HTML","stars":8270,"forks":755,"topics":["bloom","chinese-nlp","gpt-evaluation","gpt-q","instruct-finetune","instruct-gpt","instruction-set","llama","lora","open-models"],"archived":false,"github_pushed_at":"2024-10-16T11:38:59+00:00","maintenance_label":"Dormant","stars_delta_30d":-10,"url":"https://www.graphcanon.com/tools/lianjiatech-belle","markdown_url":"https://www.graphcanon.com/tools/lianjiatech-belle.md","api_url":"https://www.graphcanon.com/api/graphcanon/tools/lianjiatech-belle","graph_url":"https://www.graphcanon.com/api/graphcanon/graph?tool=lianjiatech-belle","description":"BELLE: Be Everyone's Large Language model Engine（开源中文对话大模型）","homepage_url":null,"license":"Apache-2.0","open_issues":106,"watchers":107,"ai_summary":"BELLE aims to promote the development of the open-source community for Chinese dialogue large models, focusing on making high-quality instruction-following language models accessible to everyone. The project provides instruction training data, models, training code, and evaluates the impact of different training data and algorithms on model performance.","readme_excerpt":"## <img src=\"assets/belle_logo.png\" style=\"vertical-align: middle; width: 35px;\"> BELLE: Be Everyone's Large Language model Engine\n\n*Read this in [English](README_en.md).*\n\n<div align=\"center\">\n\n<a href=\"https://github.com/LianjiaTech/BELLE/stargazers\"></a>\n\n\n\n\n\n</div>\n\n本项目的目标是促进中文对话大模型开源社区的发展，愿景是成为能够帮到每一个人的LLM Engine。\n\n相比如何做好大语言模型的预训练，BELLE更关注如何在开源预训练大语言模型的基础上，帮助每一个人都能够得到一个属于自己的、效果尽可能好的具有指令表现能力的语言模型，降低大语言模型、特别是中文大语言模型的研究和应用门槛。为此，BELLE项目会持续开放指令训练数据、相关模型、训练代码、应用场景等，也会持续评估不同训练数据、训练算法等对模型表现的影响。BELLE针对中文做了优化，模型调优仅使用由ChatGPT生产的数据（不包含任何其他数据）。\n\n</br>\n\n## 🔄 最近更新\n* [2024/10/16] 开源[Belle-whisper-larger-v3-turbo-zh](https://huggingface.co/BELLE-2/Belle-whisper-large-v3-turbo-zh)  中文能力强化后的语音识别模型，识别精度相比whisper-large-v3-turbo相对提升24~64%，识别速度相比whisper-large-v3有7-8倍提升。\n* [2024/03/15] 更新了一篇技术报告[Dial-insight](https://arxiv.org/pdf/2403.09167.pdf)  在垂直领域场景微调大模型时，使用高质量的垂直领域数据可以在使模型的垂直领域能力增强的同时，有效的抵抗模型通用能力的坍缩。\n* [2024/03/11] 开源[Belle-whisper-larger-v3-zh](https://huggingface.co/BELLE-2/Belle-whisper-large-v3-zh)  中文能力强化后的语音识别模型，相比whisper-large-v3相对提升24~65%，特别是在高噪、混响等复杂场景下有突出表现。\n* [2024/01/16] 更新了一篇技术报告[RAISE](https://arxiv.org/pdf/2401.02777.pdf). RAISE通过实验发现构造少量的样例数据，就能有效的激发大模型，生成对话也更可控\n* [2023/12/29] 开源[Belle-whisper-larger-v2-zh](https://huggingface.co/BELLE-2/Belle-whisper-large-v2-zh)和[Belle-distilwhisper-large-v2-zh](https://huggingface.co/BELLE-2/Belle-distilwhisper-large-v2-zh)两个针对中文能力强化后的语音识别模型，方便大家在语音场景下使用大语言模型\n* [2023/11/24] 开源[BELLE-VL](https://huggingface.co/BELLE-2/BELLE-VL)多模态大语言模型，基于中文能力更强的语言模型基座来扩展模型的视觉能力，为社区提供更加灵活的选择（目前BELLE-VL最新的模型在[MME](https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models/tree/Evaluation)感知评测维度共获得**1620.10**分,超过Qwen-VL、Llava、mplug-owl）\n* [2023/10/27] 更新了一篇技术报告[DUMA](https://arxiv.org/abs/2310.18075#)，探索了对话场景下基于快慢脑架构的Agent实现方法\n* [2023/09/26] 更新了RLHF的训练代码，支持PPO和[DPO](https://arxiv.org/abs/2305.18290)训练，具体细节见：[README_RLHF.md](train/README_RLHF.md)\n* [2023/08/16] 基于原有的[train_3.5M_CN](https://huggingface.co/datasets/BelleGroup/train_3.5M_CN)数据新增了指令类别字段，共包括13个类别，具体细节见：[train_3.5M_CN_With_Category](https://huggingface.co/datasets/BELLE-2/train_3.5M_CN_With_Category)\n* [2023/08/10] 更新了基于ZeRO Inference的推理代码，详见[train/README_ZERO_INFERENCE.md](train/README_ZERO_INFERENCE.md)\n* [2023/08/07] 更新了继续预训练代码和指令微调代码，添加了flash attention 2，详见[train/README.md](train/README.md)。同时打包了运行环境，详见[train/docker/README.md](train/docker/README.md)\n* [2023/07/31] 更新了一篇技术报告[ChatHome](https://arxiv.org/abs/2307.15290)，探索了针对垂直领域时的增量预训练+指令微调的的策略方法\n* [2023/07/27] 开放[BELLE-Llama2-13B-chat-0.4M](https://huggingface.co/BELLE-2/BELLE-Llama2-13B-chat-0.4M)，在Llama-2-13B的基础上采用40万高质量的对话数据上进行训练。在[评测集](https://github.com/LianjiaTech/BELLE/blob/main/eval/eval_set.json)上的效果相比BELLE-LLaMA-EXT-13B模型有显著提升。\n* [2023/05/14] 开放[BELLE-LLaMA-EXT-13B](https://huggingface.co/BelleGroup/BELLE-LLaMA-EXT-13B)，在LLaMA-13B的基础上扩展中文词表，并在400万高质量的对话数据上进行训练。\n* [2023/05/11] [BELLE/data/10M](data/10M)中，新加350万条生成多样化指令任务数据，包括单轮和多轮对话[train_3.5M_CN](https://huggingface.co/datasets/BelleGroup/train_3.5M_CN)。\n* [2023/04/19] 开放了其中一篇论文中的的相关模型：包括在LLaMA7B基础上增量预训练扩展中文词表的模（详见[BelleGroup/BELLE-LLaMA-EXT-7B](https://huggingface.co/BelleGroup/BELLE-LLaMA-EXT-7B)），以及基于多样化开源数据训练后的LLaMA-7B模型（详见[BelleGroup/BELLE-on-Open-Datasets](https://huggingface.co/BelleGroup/BELLE-on-Open-Datasets)）。\n* [2023/04/18] 更新了train代码，详见[BELLE/train](https://github.com/LianjiaTech/BELLE/tree/main/train)，集成了Deepspeed-Chat，提供了相关的docker\n* [2023/04/18] 更新了[两篇最新论文工作](#📑-研究报告)，对比了不同方式产生的训练数据、不同训练方法（LoRA, finetune)对效果的影响\n* [2023/04/12] 发布了[ChatBELLE App](chat/README.md)，基于[llama.cpp](https://github.com/ggerganov/llama.cpp)和[Flutter](https://flutter.dev/)，实现跨平台的BELLE-7B离线模型实时交互。\n* [2023/04/11] 更新了一个人工精校的eval集合，大约一千多条\n* [2023/04/08] [BELLE/data/10M](data/10M)中，新加40万条生成的给定角色的多轮对话[Generated Chat](https://huggingface.co/datasets/BelleGroup/generated_chat_0.4M)，新加200万条生成多样化指令任务数据[train_2M_CN](https://huggingface.co/datasets/BelleGroup/train_2M_CN)。\n\n</br>\n  \n\n下图是一个可以使用App在设备端本地运行4bit量化的BELLE-7B模型，在M1 Max CPU上实时运行的效果（未加速）。Ap","github_created_at":"2023-03-17T09:44:11+00:00","created_at":"2026-07-11T23:08:20.200274+00:00","updated_at":"2026-09-20T04:29:15.101832+00:00","categories":[{"slug":"computer-vision","name":"Computer Vision","url":"https://www.graphcanon.com/categories/computer-vision","markdown_url":"https://www.graphcanon.com/categories/computer-vision.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/computer-vision"},{"slug":"evaluation-observability","name":"Evaluation & Observability","url":"https://www.graphcanon.com/categories/evaluation-observability","markdown_url":"https://www.graphcanon.com/categories/evaluation-observability.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/evaluation-observability"},{"slug":"llm-frameworks","name":"LLM Frameworks","url":"https://www.graphcanon.com/categories/llm-frameworks","markdown_url":"https://www.graphcanon.com/categories/llm-frameworks.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/llm-frameworks"},{"slug":"model-training","name":"Model Training","url":"https://www.graphcanon.com/categories/model-training","markdown_url":"https://www.graphcanon.com/categories/model-training.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/model-training"},{"slug":"speech-audio","name":"Speech & Audio","url":"https://www.graphcanon.com/categories/speech-audio","markdown_url":"https://www.graphcanon.com/categories/speech-audio.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/speech-audio"}],"tags":[{"slug":"bloom","name":"bloom"},{"slug":"chinese-nlp","name":"chinese-nlp"},{"slug":"gpt-evaluation","name":"gpt-evaluation"},{"slug":"gpt-q","name":"gpt-q"},{"slug":"instruct-finetune","name":"instruct-finetune"},{"slug":"instruct-gpt","name":"instruct-gpt"},{"slug":"instruction-set","name":"instruction-set"},{"slug":"llama","name":"llama"}],"trust":{"provenance":{"is_fork":false,"github_id":615256902,"owner_type":"Organization","methodology":"github_public_v1","parent_repo":null,"near_duplicate_slugs":[]},"computed_at":"2026-09-18T08:14:05.650Z","maintenance":{"label":"Dormant","score":18,"methodology":"github_public_v1","releases_90d":0,"days_since_push":701,"last_release_at":"2023-04-12T10:53:31Z","stars_delta_30d":-10,"open_issues_delta_30d":0},"security_summary":{"status":"ok","scanner":"osv@v1","low_count":0,"high_count":0,"last_scan_at":"2026-09-18T08:14:07.876Z","medium_count":0,"scan_profile":"deps","critical_count":0}},"capability_facts":{"scan":{"source":"repo_scan","observed_at":"2026-09-18T08:14:04.543Z"},"languages":{"value":["html"],"source":"github.language","observed_at":"2026-09-18T08:14:04.543Z"},"license_spdx":{"value":"Apache-2.0","source":"github.license","observed_at":"2026-09-18T08:14:04.543Z"}},"decision_facts":{"hosting":null,"pricing":null,"requirements":{"notes":["BELLE requires Docker for running the environment.","Ensure you have sufficient computational resources for training and inference."],"min_ram_gb":8,"requires_docker":true},"constraints":{"min_ram_gb":8,"requires_docker":true},"when_to_use":["Use BELLE when you need a Chinese-specific large language model that has been optimized for instruction-following tasks.","Choose BELLE if you are working on projects that require high-quality Chinese dialogue capabilities, such as voice recognition and multi-modal applications.","Consider BELLE for research or applications where you need to evaluate the impact of different training data and algorithms on model performance."],"when_not_to_use":["Avoid BELLE if your project requires a language model that is not focused on Chinese, as it is specifically optimized for the Chinese language.","Do not use BELLE if you need a model that is trained on a diverse set of data beyond Chinese dialogue, as it primarily uses data generated by ChatGPT.","BELLE may not be suitable if you require a model that has not been fine-tuned on high-quality dialogue data, as it emphasizes instruction-following capabilities."],"source":"enrich:decision_facts","observed_at":"2026-09-18T09:44:45.116Z"},"constraint_facets":{"min_ram_gb":8,"requires_docker":true},"decision_summary":[{"label":"Requirements","value":"Min 8 GB RAM; Requires Docker; BELLE requires Docker for running the environment.; Ensure you have sufficient computational resources for training and inference."},{"label":"Adopt for","value":"BELLE is an open-source Chinese dialogue large model project that focuses on making high-quality instruction-following language models accessible to everyone. It provides instruction training data, models, training code,"},{"label":"License detail","value":"BELLE is licensed under Apache-2.0, allowing for free use, modification, and distribution."}]}}