add Qwen2-7B-Instruct_lora

2024-09-14 18:34:31 +08:00 · 2024-09-14 18:34:31 +08:00 · 198600a774
commit 198600a774
parent 8438cdfc26
5 changed files with 265 additions and 36 deletions
--- a/.gitignore
+++ b/.gitignore
@ -6,6 +6,8 @@ data/
 pdf/
 .idea/
 logs/
+.vscode/
+work_dirs/

 # *.jsonl
 # *.json
--- a/README.md
+++ b/README.md
@ -59,8 +59,9 @@
 |      ChatGLM3_6B      |   LoRA   |   [chatglm3_6b_lora_alpaca_e3.py](./xtuner_config/chatglm3_6b_lora_alpaca_e3.py)  | |
 | DeepSeek MoE_16B_chat |  QLoRA   |  [deepseek_moe_16b_chat_qlora_oasst1_e3.py](./xtuner_config/deepseek_moe_16b_chat_qlora_oasst1_e3.py)    | |
 | Mixtral 8x7B_instruct |  QLoRA   | [mixtral_8x7b_instruct_qlora_oasst1_e3.py](./xtuner_config/mixtral_8x7b_instruct_qlora_oasst1_e3.py)    | |
-| LLaMA3_8b_instruct    |  QLoRA   | [aiwei_llama3_8b_instruct_qlora_e3.py](./xtuner_config/aiwei_llama3_8b_instruct_qlora_e3.py) | [OpenXLab](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM-LLaMA3_8b_instruct_aiwei/tree/main), [ModelScope](https://modelscope.cn/models/aJupyter/EmoLLM-LLaMA3_8b_instruct_aiwei/files) |
-| LLaMA3_8b_instruct    |  QLoRA   | [llama3_8b_instruct_qlora_alpaca_e3_M_ruozhi_scM.py](./xtuner_config/llama3_8b_instruct_qlora_alpaca_e3_M_ruozhi_scM.py)    |[OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM-Llama3-8B-Instruct3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLM-Llama3-8B-Instruct3.0/summary) |
+| LLaMA3_8B_instruct    |  QLoRA   | [aiwei_llama3_8b_instruct_qlora_e3.py](./xtuner_config/aiwei_llama3_8b_instruct_qlora_e3.py) | [OpenXLab](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM-LLaMA3_8b_instruct_aiwei/tree/main), [ModelScope](https://modelscope.cn/models/aJupyter/EmoLLM-LLaMA3_8b_instruct_aiwei/files) |
+| LLaMA3_8B_instruct    |  QLoRA   | [llama3_8b_instruct_qlora_alpaca_e3_M_ruozhi_scM.py](./xtuner_config/llama3_8b_instruct_qlora_alpaca_e3_M_ruozhi_scM.py)    |[OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM-Llama3-8B-Instruct3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLM-Llama3-8B-Instruct3.0/summary) |
+| Qwen2-7B-Instruct    |  LoRA   | [Qwen2-7B-Instruct_lora.py](./xtuner_config/Qwen2-7B-Instruct_lora.py)    |[ModelScope](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen2-7B-Instruct_lora/) |
 |          ……           |    ……    |                                                    ……                                                    | …… |

 </div>
@ -100,6 +101,8 @@
 </table>

 ## 🎇最近更新
+- 【2024.09.14】基于Qwen2-7B-Instruct模型的Lora微调模型开源，微调配置文件地址：[Qwen2-7B-Instruct_lora.py](./xtuner_config/Qwen2-7B-Instruct_lora.py) ，模型权重链接：[ModelScope](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen2-7B-Instruct_lora/)
+- 【2024.08】基于GLM4-9B-chat微调Lora模型开源（基于LLaMA-Factory），详情见[微调教程](./doc/GLM-4-9B-chat%20Lora%20微调（llama-factory）.md) ，模型权重链接：[ModelScope](https://www.modelscope.cn/models/wwewwt/EmoLLM-glm-4-9b-chat/summary)
 - 【2024.07.16】欢迎大家体验 EmoLLM V3.0 ，该模型是基于InternLM2.5-7B-Chat模型的全量微调，微调配置文件地址：[internlm2_5_chat_7b_full.py](./xtuner_config/internlm2_5_chat_7b_full.py) ，模型权重链接：[OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM_V3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLMV3.0) ，WebDemo地址： [OpenXLab apps](https://openxlab.org.cn/apps/detail/chg0901/EmoLLMV3.0), [配套全量微调知乎教程](https://zhuanlan.zhihu.com/p/708931911)。
 - 【2024.07】欢迎大家使用稳定版 EmoLLM V2.0 进行日常使用和学术研究，模型权重链接：[OpenXLab](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full/tree/main)。
 - 【2024.07】新增基于InternLM2_5_7B_chat[微调配置](./xtuner_config/internlm2_5_chat_7b_qlora_oasst1_e3.py)、模型文件发布在 [ModelScope](https://www.modelscope.cn/models/z342994309/emollm_interlm2_5/)。
@ -117,15 +120,14 @@
 - 【2024.03.11】 **EmoLLM V2.0 相比 EmoLLM V1.0 全面提升，已超越 Role-playing ChatGPT 在心理咨询任务上的能力！**[点击体验EmoLLM V2.0](https://openxlab.org.cn/apps/detail/Farewell1/EmoLLMV2.0)，更新[数据集统计及详细信息](./datasets/)、[路线图](./assets/Roadmap_ZH.png)
 - 【2024.03.09】 新增并发功能加速 [QA 对生成](./scripts/qa_generation/)、[RAG pipeline](./rag/)
 - 【2024.03.03】 [基于InternLM2-7B-chat全量微调版本EmoLLM V2.0开源](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full)，需要两块A100*80G，更新专业评估，详见[evaluate](./evaluate/)，更新基于PaddleOCR的PDF转txt工具脚本，详见[scripts](./scripts/)
+
+<details>
+<summary>查看更多</summary>
 - 【2024.02.29】更新客观评估计算，详见[evaluate](./evaluate/)，更新一系列数据集，详见[datasets](./datasets/)
 - 【2024.02.27】更新英文readme和一系列数据集（舔狗和单轮对话）
 - 【2024.02.23】推出基于InternLM2_7B_chat_qlora的 `温柔御姐心理医生艾薇`，[点击获取模型权重](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_aiwei)，[配置文件](xtuner_config/aiwei-internlm2_chat_7b_qlora.py)，[在线体验链接](https://openxlab.org.cn/apps/detail/ajupyter/EmoLLM-aiwei)
 - 【2024.02.23】更新[若干微调配置](/xtuner_config/)，新增 [data_pro.json](/datasets/data_pro.json)（数量更多、场景更全、更丰富）和 [aiwei.json](/datasets/aiwei.json)（温柔御姐角色扮演专用，带有Emoji表情），即将推出 `温柔御姐心理医生艾薇`
 - 【2024.02.18】 [基于Qwen1_5-0_5B-Chat全量微调版本开源](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen1_5-0_5B-Chat_full_sft/summary)，算力有限的道友可以玩起来~
-
-<details>
-<summary>查看更多</summary>
-
 - 【2024.02.06】 EmoLLM在[**Openxlab** ](https://openxlab.org.cn/models/detail/jujimeizuo/EmoLLM_Model) 平台下载量高达18.7k，欢迎大家体验！

 <p align="center"> 
@ -185,7 +187,7 @@
          - [使用指南](#使用指南)
    - [🍪快速体验](#快速体验)
    - [📌数据构建](#数据构建)
-    - [🎨微调指南](#微调指南)
+    - [🎨增量预训练、微调指南](#增量预训练微调指南)
    - [🔧部署指南](#部署指南)
    - [⚙RAG(检索增强生成)](#rag检索增强生成)
    - [🎓评测指南](#评测指南)
@ -204,6 +206,7 @@
 ###### 开发前的配置要求

 - 硬件：A100 40G（仅针对InternLM2_7B_chat+qlora微调+deepspeed zero2优化）
+- todo：发布更多硬件消耗细节

 ###### 使用指南

@ -216,7 +219,7 @@ git clone https://github.com/SmartFlowAI/EmoLLM.git
 2. 依次阅读或者选择感兴趣的部分阅读：
   - [快速体验](#快速体验)
   - [数据构建](#数据构建)
-   - [微调指南](#微调指南)
+   - [增量预训练、微调指南](#增量预训练微调指南)
   - [部署指南](#部署指南)
   - [RAG](#rag检索增强生成)
   - [评测指南](#评测指南)
@ -230,19 +233,21 @@ git clone https://github.com/SmartFlowAI/EmoLLM.git


 ### 📌数据构建
-
 - 请阅读[数据构建指南](generate_data/tutorial.md)查阅
-
 - 微调用到的数据集见[datasets](datasets/data.json)

-### 🎨微调指南
-
-详见[微调指南](xtuner_config/README.md)
+### 🎨增量预训练、微调指南
+- 增量预训练详见[增量预训练指南](./xtuner_config/pt/README.md)
+- 【基于xtuner】全量、LoRA、QLoRA微调详见[微调指南](./xtuner_config/README.md)
+- 【基于ms-swift】全量、LoRA、QLoRA微调详见[微调指南](./swift/README.md)
+- 【基于LLaMA-Factory】全量、LoRA、QLoRA微调详见[微调指南](./doc/GLM-4-9B-chat%20Lora%20微调（llama-factory）.md)
+- todo：待更新DPO训练

 ### 🔧部署指南

 - Demo部署：详见[部署指南](demo/README.md)
 - 基于[LMDeploy](https://github.com/InternLM/lmdeploy/)的量化部署：详见[deploy](./deploy/lmdeploy.md)
+- todo: 基于VLLM部署指南

 ### ⚙RAG(检索增强生成)

@ -257,13 +262,14 @@ git clone https://github.com/SmartFlowAI/EmoLLM.git

 ### 使用到的框架

- [Xtuner](https://github.com/InternLM/xtuner)：用于微调
+- [xtuner](https://github.com/InternLM/xtuner)：用于微调
 - [Transformers](https://github.com/huggingface/transformers)
 - [Pytorch](https://pytorch.org/)
 - [LMDeploy](https://github.com/InternLM/lmdeploy/)：用于量化部署
 - [Stremlit](https://streamlit.io/)：用于构建Demo
 - [DeepSpeed](https://github.com/microsoft/DeepSpeed)：并行训练
- …
+- [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory/blob/main)：训练框架
+- [ms-swift](https://github.com/modelscope/ms-swift)：训练框架

 #### 如何参与本项目

--- a/README_EN.md
+++ b/README_EN.md
@ -63,6 +63,7 @@
 | Mixtral 8x7B_instruct |  QLoRA   | [mixtral_8x7b_instruct_qlora_oasst1_e3.py](./xtuner_config/mixtral_8x7b_instruct_qlora_oasst1_e3.py)    | |
 | LLaMA3_8b_instruct    |  QLoRA   | [aiwei_llama3_8b_instruct_qlora_e3.py](./xtuner_config/aiwei_llama3_8b_instruct_qlora_e3.py) | [OpenXLab](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM-LLaMA3_8b_instruct_aiwei/tree/main), [ModelScope](https://modelscope.cn/models/aJupyter/EmoLLM-LLaMA3_8b_instruct_aiwei/files) |
 | LLaMA3_8b_instruct    |  QLoRA   | [llama3_8b_instruct_qlora_alpaca_e3_M_ruozhi_scM.py](./xtuner_config/llama3_8b_instruct_qlora_alpaca_e3_M_ruozhi_scM.py)    |[OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM-Llama3-8B-Instruct3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLM-Llama3-8B-Instruct3.0/summary) |
+| Qwen2-7B-Instruct    |  LoRA   | [Qwen2-7B-Instruct_lora.py](./xtuner_config/Qwen2-7B-Instruct_lora.py)    |[ModelScope](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen2-7B-Instruct_lora/) |
 |          ……           |    ……    |                                                    ……                                                    | …… |


@ -104,11 +105,13 @@ The Model aims to fully understand and promote the mental health of individuals,
 </table>

 ## Recent Updates
- 【2024.07.16】 Welcome everyone to experience EmoLLM V3.0. This model is a fully fine-tuned version based on the InternLM2.5-7B-Chat model. The fine-tuning configuration file can be found at: [internlm2_5_chat_7b_full.py](./xtuner_config/internlm2_5_chat_7b_full.py). Model weights are available at: [OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM_V3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLMV3.0). WebDemo is available at: [OpenXLab apps](https://openxlab.org.cn/apps/detail/chg0901/EmoLLMV3.0), [Full fine-tuning tutorial on Zhihu](https://zhuanlan.zhihu.com/p/708931911).
- 【2024.07】Welcome to use the stable version of EmoLLM V2.0 for daily use and academic research. Model weight link: [OpenXLab](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full/tree/main).
- 【2024.07】Added InternLM2_5_7B_chat[fine-tuning configuration](./xtuner_config/internlm2_5_chat_7b_qlora_oasst1_e3.py)、model file [ModelScope](https://www.modelscope.cn/models/z342994309/emollm_interlm2_5/)。
- 【2024.06】 Added [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory)[GLM4-9B-chat fine-tuning guide](./doc/GLM-4-9B-chat%20Lora%20微调（llama-factory）.md), added [swift-based fine-tuning guide](./swift/), the paper [ESC-Eval: Evaluating Emotion Support Conversations in Large Language Models](https://arxiv.org/abs/2406.14952) cited EmoLLM and EmoLLM achieved good results.
- 【2024.05.28】The multi-turn dialogue dataset **CPsyCunD** and **professional evaluation method** used by EmoLLM have been released. For details, please see the 2024 ACL findings[《CPsyCoun: A Report-based Multi-turn Dialogue Reconstruction and Evaluation Framework for Chinese Psychological Counseling》](https://arxiv.org/abs/2405.16433)!
+- [2024.09.14] The Lora fine-tuned model based on the Qwen2-7B-Instruct model is open-sourced. Fine-tuning configuration file address: [Qwen2-7B-Instruct_lora.py](./xtuner_config/Qwen2-7B-Instruct_lora.py), model weight link: [ModelScope](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen2-7B-Instruct_lora/)
+- [2024.08] The Lora fine-tuned model based on GLM4-9B-chat is open-sourced (based on Llama-factory). For details, see [Fine-tuning Tutorial](./doc/GLM-4-9B-chat%20Lora%20微调（llama-factory）.md), model weight link: [ModelScope](https://www.modelscope.cn/models/wwewwt/EmoLLM-glm-4-9b-chat/summary)
+- [2024.07.16] Welcome everyone to experience EmoLLM V3.0. This model is a fully fine-tuned version based on the InternLM2.5-7B-Chat model. The fine-tuning configuration file can be found at: [internlm2_5_chat_7b_full.py](./xtuner_config/internlm2_5_chat_7b_full.py). Model weights are available at: [OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM_V3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLMV3.0). WebDemo is available at: [OpenXLab apps](https://openxlab.org.cn/apps/detail/chg0901/EmoLLMV3.0), [Full fine-tuning tutorial on Zhihu](https://zhuanlan.zhihu.com/p/708931911).
+- [2024.07] Welcome to use the stable version of EmoLLM V2.0 for daily use and academic research. Model weight link: [OpenXLab](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full/tree/main).
+- [2024.07] Added InternLM2_5_7B_chat[fine-tuning configuration](./xtuner_config/internlm2_5_chat_7b_qlora_oasst1_e3.py)、model file [ModelScope](https://www.modelscope.cn/models/z342994309/emollm_interlm2_5/)。
+- [2024.06] Added [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory)[GLM4-9B-chat fine-tuning guide](./doc/GLM-4-9B-chat%20Lora%20微调（llama-factory）.md), added [swift-based fine-tuning guide](./swift/), the paper [ESC-Eval: Evaluating Emotion Support Conversations in Large Language Models](https://arxiv.org/abs/2406.14952) cited EmoLLM and EmoLLM achieved good results.
+- [2024.05.28] The multi-turn dialogue dataset **CPsyCunD** and **professional evaluation method** used by EmoLLM have been released. For details, please see the 2024 ACL findings[《CPsyCoun: A Report-based Multi-turn Dialogue Reconstruction and Evaluation Framework for Chinese Psychological Counseling》](https://arxiv.org/abs/2405.16433)!
 - [2024.05.08] EmoLLM**Daddy-like BF V0.1** is public now in [1. **Baidu AppBuilder**](https://appbuilder.baidu.com/s/4cLyw) and [2. **OpenXLab**](https://openxlab.org.cn/apps/detail/chg0901/EmoLLM3.0_Gradio_Llama3-8B-Instruct3.0), welcome to like and add it to your collections!
 - [2024.05.07] [Incremental Pre-training Guide](xtuner_config/pt/README.md)
 - [2024.05.04] [EmoLLM3.0 OpenXLab Demo](https://st-app-center-006861-9746-jlroxvg.openxlab.space/) based on LLaMA3_8b_instruct is available now ([restart link]((https://openxlab.org.cn/apps/detail/chg0901/EmoLLM-Llama3-8B-Instruct3.0))), [LLAMA3 fine-tuning guide](xtuner_config/README_llama3_8b_instruct_qlora_alpaca_e3_M.md) is updated, LLaMA3_8b_instruct-8B QLoRA fine-tuning model EmoLLM3.0 weights are released on [**OpenXLab**](https://openxlab.org.cn/models/detail/chg0901/EmoLLM-Llama3-8B-Instruct3.0) and [**ModelScope**](https://modelscope.cn/models/chg0901/EmoLLM-Llama3-8B-Instruct3.0/summary) platforms
@ -122,6 +125,10 @@ The Model aims to fully understand and promote the mental health of individuals,
 - [2024.03.11] **EmoLLM V2.0 is greatly improved in all scores compared to EmoLLM V1.0. Surpasses the performance of Role-playing ChatGPT on counseling tasks!** [Click to experience EmoLLM V2.0](https://openxlab.org.cn/apps/detail/Farewell1/EmoLLMV2.0), update [dataset statistics and details](./datasets/), [Roadmap](./assets/Roadmap_ZH.png)
 - [2024.03.09] Add concurrency acceleration [QA pair generation](./scripts/qa_generation/), [RAG pipeline](./rag/)
 - [2024.03.03] [Based on InternLM2-7B-chat full fine-tuned version EmoLLM V2.0 open sourced](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full), need two A100*80G, update professional evaluation, see [evaluate](./evaluate/), update PaddleOCR-based PDF to txt tool scripts, see [scripts](./scripts/).
+
+
+<details>
+<summary>View More</summary>
 - [2024.02.29] Updated objective assessment calculations, see [evaluate](./evaluate/) for details. A series of datasets have also been updated, see [datasets](./datasets/) for details.
 - [2024.02.27] Updated English README and a series of datasets (licking dogs and one-round dialogue)
 - [2024.02.23]The "Gentle Lady Psychologist Ai Wei" based on InternLM2_7B_chat_qlora was launched. [Click here to obtain the model weights](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_aiwei), [configuration file](xtuner_config/aiwei-internlm2_chat_7b_qlora.py), [online experience link](https://openxlab.org.cn/apps/detail/ajupyter/EmoLLM-aiwei)
@ -129,11 +136,7 @@ The Model aims to fully understand and promote the mental health of individuals,
 - [2024.02.23]Updated [several fine-tuning configurations](/xtuner_config/), added [data_pro.json](/datasets/data_pro.json) (more quantity, more comprehensive scenarios, richer content) and [aiwei.json](/datasets/aiwei.json) (dedicated to the gentle lady role-play, featuring Emoji expressions), the "Gentle Lady Psychologist Ai Wei" is coming soon.

 - [2024.02.18] The full fine-tuned version based on Qwen1_5-0_5B-Chat has been [open-sourced](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen1_5-0_5B-Chat_full_sft/summary). Friends with limited computational resources can now dive in and explore it.
-
-
-<details>
-<summary>View More</summary>
-
+  
 - [2024.02.06] [Open-sourced based on the Qwen1_5-0_5B-Chat full-scale fine-tuned version](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen1_5-0_5B-Chat_full_sft/summary), friends with limited computing power can start experimenting~

 <p align="center"> 
@ -187,7 +190,7 @@ The Model aims to fully understand and promote the mental health of individuals,
          - [User Guide](#user-guide)
    - [🍪Quick start](#quick-start)
    - [📌Data Construction](#data-construction)
-    - [🎨Fine-tuning Guide](#fine-tuning-guide)
+    - [🎨Incremental Pre-training and Fine-tuning Guide](#incremental-pre-training-and-fine-tuning-guide)
    - [🔧Deployment Guide](#deployment-guide)
    - [⚙RAG (Retrieval Augmented Generation)](#rag-retrieval-augmented-generation)
    - [🎓Evaluation Guide](#evaluation-guide)
@ -206,6 +209,7 @@ The Model aims to fully understand and promote the mental health of individuals,
 ###### Pre-development Configuration Requirements.

 - A100 40G (specifically for InternLM2_7B_chat + qlora fine-tuning + deepspeed zero2 optimization)
+- **[TODO]**: Publish more details about hardware consumption.

 ###### User Guide

@ -218,7 +222,7 @@ git clone https://github.com/SmartFlowAI/EmoLLM.git
 1. Read in sequence or read sections you're interested in：
   - [Quick Start](#quick-start)
   - [Data Construction](#data-construction)
-   - [Fine-tuning Guide](#fine-tuning-guide)
+   - [Fine-tuning Guide](#incremental-pre-training-and-fine-tuning-guide)
   - [Deployment Guide](#deployment-guide)
   - [RAG](#rag-retrieval-augmented-generation)
   - [Evaluation Guide](#evaluation-guide)
@ -230,19 +234,22 @@ git clone https://github.com/SmartFlowAI/EmoLLM.git
 - Quick coding: [Baby EmoLLM](quick_start/Baby_EmoLLM.ipynb)

 ### 📌Data Construction
-
 - Please read the [Data Construction Guide ](generate_data/tutorial_EN.md) for reference.
-
 - The dataset used for this fine-tuning can be found at [datasets](datasets/data.json)

-### 🎨Fine-tuning Guide
+### 🎨Incremental Pre-training and Fine-tuning Guide
+- For details on incremental pre-training, see [Incremental Pre-training Guide](./xtuner_config/pt/README.md).
+- For full-scale, LoRA, and QLoRA fine-tuning based on **xtuner**, see [Fine-tuning Guide](./xtuner_config/README_EN.md).
+- For full-scale, LoRA, and QLoRA fine-tuning based on **ms-swift**, see [Fine-tuning Guide](./swift/README_EN.md).
+- For full-scale, LoRA, and QLoRA fine-tuning based on **LLaMA-Factory**, see [Fine-tuning Guide](./doc/GLM-4-9B-chat%20Lora%20微调（llama-factory）.md).
+- **[TODO]**: Update DPO training.

-For details, see the [fine-tuning guide](xtuner_config/README_EN.md)

 ### 🔧Deployment Guide

 - Demo deployment: see [deployment guide](./demo/README_EN.md) for details.
 - Quantitative deployment based on [LMDeploy](https://github.com/InternLM/lmdeploy/): see [deploy](./deploy/lmdeploy_EN.md)
+- **[TODO]**: Deployment Guide for VLLM

 ### ⚙RAG (Retrieval Augmented Generation)

@ -263,7 +270,8 @@ For details, see the [fine-tuning guide](xtuner_config/README_EN.md)
 - [LMDeploy](https://github.com/InternLM/lmdeploy/): for quantitative deployment
 - [Stremlit](https://streamlit.io/): for building demos
 - [DeepSpeed](https://github.com/microsoft/DeepSpeed): for parallel training
- …
+- [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory/blob/main)
+- [ms-swift](https://github.com/modelscope/ms-swift)

 #### How to participate in this project

--- a/scripts/upload_modelscope.py
+++ b/scripts/upload_modelscope.py
@ -1,11 +1,11 @@

 from modelscope.hub.api import HubApi

-YOUR_ACCESS_TOKEN = '' #输入你的modelscope access token
+YOUR_ACCESS_TOKEN = ''  # 输入你的modelscope access token

 api = HubApi()
 api.login(YOUR_ACCESS_TOKEN)
 api.push_model(
-            model_id="zealot5209/EmoLLM-Scientist",  #your_name/model_id
-                model_dir="./merged" # 本地模型目录，要求目录中必须包含configuration.json
-                )
+    model_id="zealot5209/EmoLLM-Scientist",  # your_name/model_id
+    model_dir="./merged"  # 本地模型目录，要求目录中必须包含configuration.json
+)
--- a/xtuner_config/Qwen2-7B-Instruct_lora.py
+++ b/xtuner_config/Qwen2-7B-Instruct_lora.py
@ -0,0 +1,213 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import torch
+from datasets import load_dataset
+from mmengine.dataset import DefaultSampler
+from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
+                            LoggerHook, ParamSchedulerHook)
+from mmengine.optim import AmpOptimWrapper, CosineAnnealingLR, LinearLR
+from peft import LoraConfig
+from torch.optim import AdamW
+from transformers import (AutoModelForCausalLM, AutoTokenizer,
+                          BitsAndBytesConfig)
+
+from xtuner.dataset import process_hf_dataset
+from xtuner.dataset.collate_fns import default_collate_fn
+from xtuner.dataset.map_fns import alpaca_map_fn, template_map_fn_factory
+from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
+                                 VarlenAttnArgsToMessageHubHook)
+from xtuner.engine.runner import TrainLoop
+from xtuner.model import SupervisedFinetune
+from xtuner.parallel.sequence import SequenceParallelSampler
+from xtuner.utils import PROMPT_TEMPLATE, SYSTEM_TEMPLATE
+
+#######################################################################
+#                          PART 1  Settings                           #
+#######################################################################
+# Model
+pretrained_model_name_or_path = 'Qwen2-7B-Instruct'  # your model path
+use_varlen_attn = False
+
+# Data
+alpaca_en_path = '../datasets/aiwei.json'
+prompt_template = PROMPT_TEMPLATE.qwen_chat
+max_length = 1024
+pack_to_max_length = True
+
+# parallel
+sequence_parallel_size = 1
+
+# Scheduler & Optimizer
+batch_size = 8  # per_device
+accumulative_counts = 16
+accumulative_counts *= sequence_parallel_size
+dataloader_num_workers = 4
+max_epochs = 3
+optim_type = AdamW
+lr = 1e-5
+betas = (0.9, 0.999)
+weight_decay = 0
+max_norm = 1  # grad clip
+warmup_ratio = 0.03
+
+# Save
+save_steps = 100
+save_total_limit = 2  # Maximum checkpoints to keep (-1 means unlimited)
+
+# Evaluate the generation performance during the training
+evaluation_freq = 100
+SYSTEM = "现在你是一个心理专家，我有一些心理问题，请你用专业的知识帮我解决。"
+evaluation_inputs = [
+    '我压力很大', '生活没意思', "非常容易羡慕别人啊"
+]
+
+#######################################################################
+#                      PART 2  Model & Tokenizer                      #
+#######################################################################
+tokenizer = dict(
+    type=AutoTokenizer.from_pretrained,
+    pretrained_model_name_or_path=pretrained_model_name_or_path,
+    trust_remote_code=True,
+    padding_side='right')
+
+model = dict(
+    type=SupervisedFinetune,
+    use_varlen_attn=use_varlen_attn,
+    llm=dict(
+        type=AutoModelForCausalLM.from_pretrained,
+        pretrained_model_name_or_path=pretrained_model_name_or_path,
+        trust_remote_code=True,
+        torch_dtype=torch.float16,
+    ),
+    lora=dict(
+        type=LoraConfig,
+        r=32,
+        lora_alpha=16,
+        lora_dropout=0.1,
+        bias='none',
+        task_type='CAUSAL_LM'))
+
+#######################################################################
+#                      PART 3  Dataset & Dataloader                   #
+#######################################################################
+alpaca_en = dict(
+    type=process_hf_dataset,
+    dataset=dict(type=load_dataset, path='json',
+                 data_files=dict(train=alpaca_en_path)),
+    tokenizer=tokenizer,
+    max_length=max_length,
+    dataset_map_fn=None,
+    template_map_fn=dict(
+        type=template_map_fn_factory, template=prompt_template),
+    remove_unused_columns=True,
+    shuffle_before_pack=True,
+    pack_to_max_length=pack_to_max_length,
+    use_varlen_attn=use_varlen_attn)
+
+sampler = SequenceParallelSampler \
+    if sequence_parallel_size > 1 else DefaultSampler
+
+train_dataloader = dict(
+    batch_size=batch_size,
+    num_workers=dataloader_num_workers,
+    dataset=alpaca_en,
+    sampler=dict(type=sampler, shuffle=True),
+    collate_fn=dict(type=default_collate_fn, use_varlen_attn=use_varlen_attn))
+
+#######################################################################
+#                    PART 4  Scheduler & Optimizer                    #
+#######################################################################
+# optimizer
+optim_wrapper = dict(
+    type=AmpOptimWrapper,
+    optimizer=dict(
+        type=optim_type, lr=lr, betas=betas, weight_decay=weight_decay),
+    clip_grad=dict(max_norm=max_norm, error_if_nonfinite=False),
+    accumulative_counts=accumulative_counts,
+    loss_scale='dynamic',
+    dtype='float16')
+
+# learning policy
+# More information: https://github.com/open-mmlab/mmengine/blob/main/docs/en/tutorials/param_scheduler.md  # noqa: E501
+param_scheduler = [
+    dict(
+        type=LinearLR,
+        start_factor=1e-5,
+        by_epoch=True,
+        begin=0,
+        end=warmup_ratio * max_epochs,
+        convert_to_iter_based=True),
+    dict(
+        type=CosineAnnealingLR,
+        eta_min=0.0,
+        by_epoch=True,
+        begin=warmup_ratio * max_epochs,
+        end=max_epochs,
+        convert_to_iter_based=True)
+]
+
+# train, val, test setting
+train_cfg = dict(type=TrainLoop, max_epochs=max_epochs)
+
+#######################################################################
+#                           PART 5  Runtime                           #
+#######################################################################
+# Log the dialogue periodically during the training process, optional
+custom_hooks = [
+    dict(type=DatasetInfoHook, tokenizer=tokenizer),
+    dict(
+        type=EvaluateChatHook,
+        tokenizer=tokenizer,
+        every_n_iters=evaluation_freq,
+        evaluation_inputs=evaluation_inputs,
+        system=SYSTEM,
+        prompt_template=prompt_template)
+]
+
+if use_varlen_attn:
+    custom_hooks += [dict(type=VarlenAttnArgsToMessageHubHook)]
+
+# configure default hooks
+default_hooks = dict(
+    # record the time of every iteration.
+    timer=dict(type=IterTimerHook),
+    # print log every 10 iterations.
+    logger=dict(type=LoggerHook, log_metric_by_epoch=False, interval=10),
+    # enable the parameter scheduler.
+    param_scheduler=dict(type=ParamSchedulerHook),
+    # save checkpoint per `save_steps`.
+    checkpoint=dict(
+        type=CheckpointHook,
+        by_epoch=False,
+        interval=save_steps,
+        max_keep_ckpts=save_total_limit),
+    # set sampler seed in distributed evrionment.
+    sampler_seed=dict(type=DistSamplerSeedHook),
+)
+
+# configure environment
+env_cfg = dict(
+    # whether to enable cudnn benchmark
+    cudnn_benchmark=False,
+    # set multi process parameters
+    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
+    # set distributed parameters
+    dist_cfg=dict(backend='nccl'),
+)
+
+# set visualizer
+visualizer = None
+
+# set log level
+log_level = 'INFO'
+
+# load from which checkpoint
+load_from = None
+
+# whether to resume training from the loaded checkpoint
+resume = False
+
+# Defaults to use random seed and disable `deterministic`
+randomness = dict(seed=None, deterministic=False)
+
+# set log processor
+log_processor = dict(by_epoch=False)