docs: Add Chinese readme
This commit is contained in:
KMnO4-zx
@@ -1,5 +1,9 @@
|
||||
# EasyDistill: Easy Knowledge Distillation for Large Language Models
|
||||
|
||||
<div align="center">
|
||||
[中文](./README_zh.md) | [English](./README.md)
|
||||
</div>
|
||||
|
||||
Introducing **EasyDistill**, a pioneering toolkit on knowledge distillation (KD) for large language models (LLMs). With the growing complexity and size of LLMs, **EasyDistill** offers a versatile and user-friendly platform to streamline the KD process, supporting both black-box and white-box methodologies. It facilitates efficient model training, enabling smaller models to emulate the performance of larger ones without compromising accuracy. **EasyDistill** boasts an extensive range of features, including data synthesis, supervised fine-tuning, ranking optimization, and reinforcement learning, all tailored for various KD scenarios. Designed to accommodate both System 1 (fast, intuitive) and System 2 (slow, analytical) cognitive models, the toolkit is modular and easy to use, with a simple command-line interface guiding users. Beyond academic exploration, **EasyDistill** anchors practical industrial solutions, offering robust distilled models and open-source datasets, while also showcasing seamless integration with Alibaba Cloud’s AI platform, PAI. Committed to bridging theoretical advancements with practical needs, **EasyDistill** empowers the NLP community, making state-of-the-art KD strategies accessible to researchers and industry practitioners alike.
|
||||
|
||||
# Technical Articles
|
||||
|
||||
193
README_zh.md
Normal file
193
README_zh.md
Normal file
@@ -0,0 +1,193 @@
|
||||
# EasyDistill: Easy Knowledge Distillation for Large Language Models
|
||||
|
||||
<div align="center">
|
||||
[中文](./README_zh.md) | [English](./README.md)
|
||||
</div>
|
||||
|
||||
**EasyDistill** 是一个专为大语言模型（LLMs）知识蒸馏（KD）而设计的开创性工具包。随着大语言模型复杂性和规模的不断增长，**EasyDistill** 提供了一个多功能且用户友好的平台来简化知识蒸馏过程，支持黑盒和白盒两种方法。它促进高效的模型训练，使较小的模型能够在不损失准确性的情况下模拟较大模型的性能。**EasyDistill** 拥有广泛的功能特性，包括数据合成、监督微调、排序优化和强化学习，所有这些都针对各种知识蒸馏场景进行了定制。该工具包设计用于适应系统1（快速、直觉）和系统2（缓慢、分析）认知模型，具有模块化和易于使用的特点，配备简单的命令行界面来指导用户。除了学术探索之外，**EasyDistill** 还锚定实际的工业解决方案，提供强大的蒸馏模型和开源数据集，同时展示与阿里云AI平台PAI的无缝集成。致力于连接理论进步与实际需求，**EasyDistill** 赋能NLP社区，使最先进的知识蒸馏策略对研究人员和行业从业者都变得可及。
|
||||
|
||||
# 技术文章
|
||||
|
||||
我们有一系列关于EasyDistill功能的技术文章。
|
||||
|
||||
- [人工智能平台 PAI DistilQwen2.5-DS3-0324发布：知识蒸馏+快思考=更高效解决推理难题](https://developer.aliyun.com/article/1661734)
|
||||
- [DistilQwen2.5-R1发布：知识蒸馏助推小模型深度思考](https://developer.aliyun.com/article/1659288)
|
||||
- [DistilQwen2.5发布：通义千问蒸馏小模型再升级](https://developer.aliyun.com/article/1653842)
|
||||
- [DistilQwen2：通义千问大模型的知识蒸馏实践](https://developer.aliyun.com/article/1633882)
|
||||
- [基于多轮课程学习的大语言模型蒸馏算法TAPIR](https://developer.aliyun.com/article/1635146)
|
||||
|
||||
|
||||
|
||||
## 概述
|
||||
|
||||
|
||||
|
||||
- **工具包特性**：EasyDistill提供多功能性，包括数据合成、监督微调、logits蒸馏、排序优化和针对知识蒸馏场景定制的强化学习技术。
|
||||
- **兼容性**：支持系统1（快速、直觉）和系统2（缓慢、分析）模型。
|
||||
- **用户友好**：凭借其模块化设计和简单的命令行界面，EasyDistill使知识蒸馏策略的实验和实现变得简单直接。
|
||||
- **工业集成**：整合基于知识蒸馏的解决方案，支持与阿里云人工智能平台（PAI）等平台的集成。
|
||||
|
||||
|
||||
## 快速开始
|
||||
|
||||
1. 克隆仓库：
|
||||
```bash
|
||||
git clone https://github.com/modelscope/easydistill
|
||||
cd EasyDistill
|
||||
```
|
||||
|
||||
2. 安装所需依赖：
|
||||
```bash
|
||||
python setup.py install
|
||||
```
|
||||
|
||||
3. 通过命令行界面探索EasyDistill的使用：
|
||||
```bash
|
||||
easydistill --config <config-file-path>
|
||||
```
|
||||
|
||||
配置文件表达了**EasyDistill**支持的任何知识蒸馏作业的详细设置。黑盒蒸馏配置示例如下所示：
|
||||
```json
|
||||
{
|
||||
"job_type": "kd_black_box_local",
|
||||
"dataset": {
|
||||
"instruction_path": "train.json",
|
||||
"labeled_path": "train_labeled.json",
|
||||
"template" : "chat_template/chat_template_kd.jinja",
|
||||
"seed": 42
|
||||
},
|
||||
"inference":{
|
||||
"enable_chunked_prefill": true,
|
||||
"seed": 777,
|
||||
"gpu_memory_utilization": 0.9,
|
||||
"temperature": 0.8,
|
||||
"trust_remote_code": true,
|
||||
"enforce_eager": false,
|
||||
"max_model_len": 4096,
|
||||
"max_new_tokens": 512
|
||||
},
|
||||
"models": {
|
||||
"teacher": "teacher/Qwen/Qwen2.5-7B-Instruct/",
|
||||
"student": "student/Qwen/Qwen2.5-0.5B-Instruct/"
|
||||
},
|
||||
"training": {
|
||||
"output_dir": "./result/",
|
||||
"num_train_epochs": 3,
|
||||
"per_device_train_batch_size": 1,
|
||||
"gradient_accumulation_steps": 8,
|
||||
"save_steps": 1000,
|
||||
"max_length": 512,
|
||||
"logging_steps": 1,
|
||||
"learning_rate": 2e-5,
|
||||
"weight_decay": 0.05,
|
||||
"warmup_ratio": 0.1,
|
||||
"lr_scheduler_type": "cosine"
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## DistilQwen系列
|
||||
|
||||
**DistilQwen**模型代表了从**EasyDistill**工具包衍生的强大蒸馏语言模型套件。基于知识蒸馏原理设计，DistilQwen模型在保持高性能的同时显著减少了模型大小，使其非常适合资源受限的环境。无论您是希望在工业场景中高效部署，还是寻求探索先进的知识蒸馏方法，**DistilQwen**模型都准备好以敏捷性和精确性满足多样化的应用需求。
|
||||
|
||||
|
||||
### 最新进展：自适应思维模型
|
||||
|
||||
最新的**DistilQwen**系列是**DistilQwen-ThoughtX**，与其前身相比，它展现出改进的推理能力并生成更优长度的思维链（CoTs）。该模型系列是通过利用新颖的推理冗余度（RV）和认知难度（CD）评分从创新的**OmniThought**数据集开发而来，这些评分确保模型接收到反映最优CoT输出长度和难度的丰富、高质量训练数据。**DistilQwen-ThoughtX**在开源社区中超越了其他知识蒸馏模型。**DistilQwen-ThoughtX**的性能如下所示。
|
||||
|
||||
|
||||
| **Model** | **AIME2024** | **MATH500** | **GPQA-D** | **LCB V2** | **Avg.** |
|
||||
|-----------------------------------------------|--------------|-------------|------------|------------|-----------|
|
||||
| OpenThinker-7B | 31.3 | 83.0 | 42.4 | 39.9 | 49.1 |
|
||||
| DeepSeek-R1-Distill-Qwen-7B | **57.3** | _89.6_ | 47.3 | 48.4 | 60.6 |
|
||||
| OpenThinker2-7B | 50.0 | 88.4 | _49.3_ | _55.6_ | _60.8_ |
|
||||
| **DistilQwen-ThoughtX-7B** | _56.7_ | **90.2** | **50.0** | **56.8** | **63.4** |
|
||||
| LIMO-32B | 56.7 | 86.6 | 58.1 | 60.0 | 65.3 |
|
||||
| OpenThinker-32B | 66.0 | 90.6 | 61.6 | 68.9 | 71.7 |
|
||||
| DeepSeek-R1-Distill-Qwen-32B | 74.7 | 90.0 | 62.4 | 72.3 | 74.8 |
|
||||
| OpenThinker2-32B | _76.7_ | _90.8_ | **64.1** | _72.5_ | _76.0_ |
|
||||
| Light-R1-32B | 74.7 | 90.4 | 62.0 | 56.0 | 70.7 |
|
||||
| s1.1-32B | 59.3 | 87.4 | 62.0 | 58.7 | 66.8 |
|
||||
| **DistilQwen-ThoughtX-32B** | **80.0** | **92.6** | _64.0_ | **73.4** | **77.5** |
|
||||
|
||||
**OmniThought**数据集也已公开发布。请参考数据集部分。
|
||||
|
||||
### 系统1模型
|
||||
|
||||
**DistilQwen2**是Qwen2模型的增强版本，配备了针对各种NLP任务的改进指令跟随能力。我们使用GPT-4和Qwen-max作为教师模型来生成高质量响应，平衡输入指令的任务分布。在监督微调（SFT）之后，使用DPO算法执行排序优化过程，以增强学生模型与教师模型之间的对齐。**DistilQwen2.5**模型使用黑盒和白盒知识蒸馏算法的组合进行训练。我们遵循与**DistilQwen2**生产中采用的相同指令数据处理和黑盒SFT程序。随后，应用白盒训练来完善学生从教师模型获取复杂知识的能力，特别是利用Qwen2.5-72B-Instruct作为开源教师模型。**DistilQwen2**和**DistilQwen2.5**的性能如下所示。
|
||||
|
||||
| **Model** | **AlpacaEval 2.0 (length control)** | **MT-Bench** | **MT-Bench (single)** | **IFEval (instruct-loose)** | **IFEval (strict-prompt)** |
|
||||
|------------------------------------|-------------------------------------|--------------|-----------------------|-----------------------------|----------------------------|
|
||||
| Qwen2.5-0.5B-Instruct | 2.46 | 5.49 | 6.26 | 42.81 | 30.31 |
|
||||
| **DistilQwen2.5-0.5B-Instruct** | **4.89** | **5.78** | **6.83** | **52.61** | **37.82** |
|
||||
| Qwen2-1.5B-Instruct | 5.22 | 5.85 | 6.45 | 41.37 | 28.10 |
|
||||
| **DistilQwen2-1.5B-Instruct** | **8.28** | **6.42** | **7.12** | **49.76** | **36.04** |
|
||||
| Qwen2.5-1.5B-Instruct | 6.69 | 7.09 | 7.66 | 55.40 | 40.11 |
|
||||
| **DistilQwen2.5-1.5B-Instruct** | **13.69** | **7.35** | **7.99** | **61.10** | **74.49** |
|
||||
| Qwen2.5-3B-Instruct | 17.98 | 7.92 | 8.40 | 61.18 | 74.58 |
|
||||
| **DistilQwen2.5-3B-Instruct** | **20.91** | **8.37** | **8.97** | **67.03** | **77.36** |
|
||||
| Qwen2-7B-Instruct | 24.33 | 8.27 | 8.68 | 66.67 | 52.31 |
|
||||
| **DistilQwen2-7B-Instruct** | **25.35** | **8.40** | **9.03** | **71.46** | **60.26** |
|
||||
| Qwen2.5-7B-Instruct | 31.43 | 8.52 | 8.83 | 81.53 | 72.10 |
|
||||
| **DistilQwen2.5-7B-Instruct** | **34.86** | **8.76** | **9.22** | **83.48** | **73.27** |
|
||||
|
||||
|
||||
我们已向公众发布了两个指令跟随数据集。请参考数据集部分。
|
||||
|
||||
|
||||
### 系统2模型
|
||||
|
||||
**DistilQwen2.5-R1**模型系列利用DeepSeek-R1作为教师模型。为了将较小蒸馏模型的推理能力与其内在认知能力对齐，模型使用我们的CogPO算法进一步完善，该算法优于其他训练方法。此外，我们将DeepSeek-V3-0324的快速思维推理能力转移到**DistilQwen2.5-DS3-0324**模型中。为了缩短推理过程，采用CoT简化操作符来减少**DistilQwen2.5-R1**训练数据中的token数量。结合包含DeepSeek-V3-0324的CoT蒸馏数据的重写数据集，我们开发了**DistilQwen2.5-DS3-0324**模型。**DistilQwen2.5-R1**和**DistilQwen2.5-DS3-0324**的性能如下所示。
|
||||
|
||||
| **Model** | **AIME2024** | **MATH-500** | **GPQA Diamond** | **LiveCodeBench V2** |
|
||||
|---------------------------------------|--------------|--------------|------------------|----------------------|
|
||||
| Qwen2.5-3B-Instruct | 6.67 | 62.6 | 32.83 | 11.35 |
|
||||
| **DistilQwen2.5-DS3-0324-3B** | **16.67** | **70.0** | **34.34** | **18.00** |
|
||||
| Qwen2.5-7B-Instruct | 10.0 | 73.6 | 33.30 | 30.72 |
|
||||
| **DistilQwen2.5-7B-R1** | **23.33** | **77.8** | **37.88** | **36.40** |
|
||||
| **DistilQwen2.5-DS3-0324-7B** | **43.33** | **88.4** | **42.93** | **46.38** |
|
||||
| Qwen2.5-14B-Instruct | 16.7 | 78.2 | 43.43 | 37.38 |
|
||||
| **DistilQwen2.5-14B-R1** | **26.67** | **82.6** | **45.45** | **41.49** |
|
||||
| **DistilQwen2.5-DS3-0324-14B** | **46.67** | **90.8** | **51.52** | **54.40** |
|
||||
| Qwen2.5-32B-Instruct | 16.67 | 81.4 | 45.50 | 47.36 |
|
||||
| **DistilQwen2.5-32B-R1** | **46.67** | **87.0** | **48.99** | **55.97** |
|
||||
| **DistilQwen2.5-DS3-0324-32B** | **70.00** | **93.8** | **62.12** | **65.95** |
|
||||
|
||||
所有**DistilQwen**模型都在HuggingFace和ModelScope上公开发布。
|
||||
|
||||
## 发布的数据集
|
||||
|
||||
我们还基于**EasyDistill**框架发布了几个数据集。
|
||||
|
||||
### 指令跟随数据集
|
||||
|
||||
为了帮助社区开发者在微调**DistilQwen**模型时避免灾难性遗忘，我们开源了两个数据集：**DistilQwen_100K**和**DistilQwen_1M**。这些数据集旨在为模型微调提供坚实的基础，增强对新任务的适应性，同时保持在先前任务上的性能。此外，它还可以用于在微调其他类似的大语言模型时改善指令跟随能力。这些数据集涵盖了广泛的内容，包括数学、代码、基于知识的问答、指令跟随和创意生成，总数据集大小分别为100K和1M条目。用户可以在模型微调期间将**DistilQwen_100K**和**DistilQwen_1M**或其子集与自己的数据集成，以确保出色的下游任务性能，同时保持模型的通用能力，从而保持其泛化能力。
|
||||
|
||||
|
||||
### 思维链推理数据集
|
||||
|
||||
**OmniThought**是一个大规模数据集，包含由DeepSeek-R1和QwQ-32B生成和验证的**200万**个思维链（CoT）过程。**OmniThought**中的每个CoT过程都用新颖的推理冗余度（RV）和认知难度（CD）评分进行标注，这些评分描述了CoT冗余度的适当性和模型理解这些推理过程的认知难度水平。基于我们的**OmniThought**数据集，我们进一步训练并发布了一系列高性能模型（**DistilQwen-ThoughtX-7B**和**DistilQwen-ThoughtX-32B**），特别配备了更强的推理能力和最优的CoT输出长度和难度水平。详情请参考`recipes/open_datasets`。
|
||||
|
||||
所有数据集都在HuggingFace和ModelScope上公开发布。
|
||||
|
||||
|
||||
## 参考文献
|
||||
|
||||
我们有[一篇arxiv论文](https://arxiv.org/abs/2505.20888)供您引用EasyDistill库。以下是与我们项目相关的论文。
|
||||
|
||||
- Chengyu Wang, Junbing Yan, Wenrui Cai, Yuanhao Yue, Jun Huang. EasyDistill: A Comprehensive Toolkit for Effective Knowledge Distillation of Large Language Models. arXiv preprint
|
||||
- Wenrui Cai, Chengyu Wang, Junbing Yan, Jun Huang, Xiangzhong Fang. Reasoning with OmniThought: A Large CoT Dataset with Verbosity and Cognitive Difficulty Annotations. arXiv preprint
|
||||
- Wenrui Cai, Chengyu Wang, Junbing Yan, Jun Huang, Xiangzhong Fang. Training Small Reasoning LLMs with Cognitive Preference Alignment. arXiv preprint
|
||||
- Chengyu Wang, Junbing Yan, Yuanhao Yue, Jun Huang. DistilQwen2.5: Industrial Practices of Training Distilled Open Lightweight Language Models. **ACL 2025**
|
||||
- Yuanhao Yue, Chengyu Wang, Jun Huang, Peng Wang. Building a Family of Data Augmentation Models for Low-cost LLM Fine-tuning on the Cloud. **COLING 2025**
|
||||
- Yuanhao Yue, Chengyu Wang, Jun Huang, Peng Wang. Distilling Instruction-following Abilities of Large Language Models with Task-aware Curriculum Planning. **EMNLP 2024**
|
||||
|
||||
|
||||
## 许可证
|
||||
|
||||
本项目采用[Apache许可证（版本2.0）](LICENSE)授权。该工具包还包含一些从其他仓库修改的代码，这些代码采用其他开源许可证。更多信息请参见[NOTICE](NOTICE)文件。
|
||||
|
||||
|
||||
## 参与讨论
|
||||
|
||||
我们欢迎社区伙伴协作并为开发做出贡献，欢迎加入钉钉群：117440002081 参与讨论。
|
||||
Reference in New Issue
Block a user