Feature/issue 25/create chat with repo compoent #38
Conversation
innovation64
commented
Jan 13, 2024
innovation64
commented
- refactor: refactor the code
- style: change the gradio UI
- fix: remove useless files
- fix: add config.yml
| self.logger.debug(f"Results: {results}") | ||
| chunkrecall = self.extract_and_format_documents(result) | ||
| retrieved_documents = results['documents'][0] | ||
| response = self.rag(prompt,retrieved_documents) |
There was a problem hiding this comment.
还有因为没有切块，所以每个索引的块内容都很长，不适合多个结果一块送入，会超过4096token,大概召回3个左右就 24900个左右了，所以把相关的都放进embedding recall里面了，没有再经过LLM做最后的输出。
所以目前的做法是只将 Top 1 的 rag 结果，即 md 文档中的内容给到大模型吗?
| assistant = RepoAssistant(api_key, api_base, db_path,log_file) | ||
| md_contents = assistant.json_data.extract_md_contents() | ||
| assistant.chroma_data.create_vector_store(md_contents) | ||
| GradioInterface(assistant.respond) |
There was a problem hiding this comment.
这个写法我持保留意见，我觉得当前 main 的目的是清晰地表达调用过程，但这行将以下过程进行了封装 1. 类实例化 2. 传递 RepoAssistant 类方法名 3. 将 assistant.respond 方法交由 GradioInterface 类执行，即与清晰的目的不一致。
There was a problem hiding this comment.
这四部指的是1，传入相关配置，2.获取md内容3，放到向量数据库，4启动查询UI，主要rag已经封装了，上次说的能单例化的我基本都单例开了个类
There was a problem hiding this comment.
我是针对 GradioInterface(assistant.respond) 这一行感觉 too heavy， 但只是个人意见，Git 默认是显示前四行。
repo_agent/chat_with_repo/rag.py
Outdated
| chroma_data = ChromaManager(api_key, api_base) | ||
| self.textanslys = textanslys | ||
| self.json_data = json_data | ||
| self.chroma_data = chroma_data |
There was a problem hiding this comment.
如果我的话应该会这么写：self.chroma_data = ChromaManager(api_key, api_base)
| file_handler.setLevel(log_level) | ||
| formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s') | ||
| file_handler.setFormatter(formatter) | ||
| self.logger.addHandler(file_handler) |
There was a problem hiding this comment.
在下个版本中使用 from loguru import logger 吧，是一个简化了 Logging 配置的库
There was a problem hiding this comment.
在下个版本中使用
from loguru import logger吧，是一个简化了 Logging 配置的库
好的，下次更新就换掉
