民间版中文羊驼模型 v3.2

[ymcui]

本次以功能性更新为主，包括添加指令精调脚本、LangChain支持、基于Gradio的本地Demo等。

添加了指令精调脚本，优化实验启动流程（#322）

• 添加了指令精调代码 run_clm_sft_with_peft.py
• 优化了预训练和精调启动流程，现在可以更方便地运行实验了
  • 添加了预训练启动脚本run_pt.sh和精调启动脚本run_sft.sh
  • 添加了预训练notebook和指令精调notebook，可直接在jupyter-notebook或Colab中执行
• 详细使用方法请参考Wiki中的相关教程

添加了LangChain使用示例（#338）

• Chinese-Alpaca可以轻松地与LLM应用程序开发框架LangChain结合进行二次开发
• 添加了使用Chinese-Alpaca和LangChain进行 检索问答（Retrieval QA） 和 摘要生成（Summarization） 的示例
• 详细使用方法请参考Wiki中的LangChain教程

添加了基于Gradio的本地Web Demo（#300）

• 在本地浏览器中使用原生的🤗transformers进行推理，并且支持多轮对话
• 请参考wiki中的使用Transformers推理

感谢@sunyuhan19981208 贡献的PR。

其他更新

• 🤗transformers推理脚本inference_hf.py现在支持使用多GPU进行推理（Support multi-gpu transformers inference #272）
• 修复了alpaca_data_zh_51k.json指令精调数据
• 修复了merge_llama_with_chinese_lora.py保存多个分块的一个问题

For English release note, please refer to Discussion.

---

This discussion was created from the release 民间版中文羊驼模型 v3.2.

[ymcui]

English Release Note v3.2

This update primarily focuses on functional enhancements, including the addition of command fine-tuning scripts, LangChain support, and a local demo based on Gradio.

Added command fine-tuning scripts, optimized experiment startup process (#322)

• Command fine-tuning code run_clm_sft_with_peft.py has been added.
• The pre-training and fine-tuning startup process has been optimized, making running experiments more convenient.
  • Pre-training startup script run_pt.sh and fine-tuning startup script run_sft.sh have been added.
  • A pre-training notebook and a instruction fine-tuning notebook have been added, which can be run directly in jupyter-notebook or Colab.
• For detailed usage, please refer to the tutorial in the Wiki.

Added LangChain usage example (#338)

• Chinese-Alpaca can be easily integrated with the LLM application development framework LangChain for secondary development.
• Added examples of Retrieval QA and Summarization using Chinese-Alpaca and LangChain.
• For detailed usage, please refer to the LangChain tutorial in the Wiki.

Added a local Web Demo based on Gradio (#300)

• Run inference with native 🤗transformers in the local browser, also supporting multi-round dialogues.
• Please refer to Using Transformers for Inference in the wiki.

Thanks to @sunyuhan19981208 for the contributed PR.

Other updates

• The 🤗transformers inference script inference_hf.py now supports multi-GPU inference (Support multi-gpu transformers inference #272).
• Fixed instruction fine-tuning data alpaca_data_zh_51k.json.
• Fixed an issue with saving multiple shards in merge_llama_with_chinese_lora.py.