替换模型

[11Czh]

想问一下，该RAG框架中各个模块（嵌入模型或者llm）可以替换成微调之后的么？

[wzdavid]

可以替换。嵌入模型放在localmodels。微调后的LLM可以通过配置API访问。

[11Czh]

好的，非常感谢您的解答，按照你说的方法，我试一下 回忆、是幸福的延续 つ ***@***.***  

…

------------------ 原始邮件 ------------------ 发件人: "wzdavid/ThinkRAG" ***@***.***>; 发送时间: 2025年3月5日(星期三) 上午9:07 ***@***.***>; 抄送: "回忆、是幸福的延续 ***@***.******@***.***>; 主题: Re: [wzdavid/ThinkRAG] 替换模型 (Issue #20) 可以替换。嵌入模型放在localmodels。微调后的LLM可以通过配置API访问。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***> wzdavid left a comment (wzdavid/ThinkRAG#20) 可以替换。嵌入模型放在localmodels。微调后的LLM可以通过配置API访问。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[11Czh]

您好，针对Thinkrag这个开源的框架，我想通过微调模型，来测试微调前后的一个效果，想通过我们这个框架实现批量处理输入的问题，生成答案来进行微调前后的一个效果，想问一下我们这个框架可以实现么，不仅仅在前端进行问答，期待您的回复 回忆、是幸福的延续 つ ***@***.***  

[wzdavid]

ThinkRAG是基于LlamaIndex编写的。LlamaIndex有评估模块，可以评估RAG的效果，也许适合你的场景，但ThinkRAG还没有使用它。
https://docs.llamaindex.ai/en/stable/understanding/evaluating/evaluating/

[richlysakowski]

I translated all comments in this thread to English so I can understand what progress is being mdade.

Here are the translated comments from Issue #20 on the GitHub repository wzdavid/ThinkRAG:

---

ORIGINAL POST by 11Czh:

I would like to ask if the various modules (embedding model or LLM) in this RAG framework can be replaced with fine-tuned ones?

1. Comment by wzdavid (March 5, 2025):

   It can be replaced. The embedding model is placed in localmodels. The fine-tuned LLM can be accessed through a configured API.
   View Comment on GitHub

2. Comment by 11Czh (March 5, 2025):

   Okay, thank you very much for your explanation. I will try it according to your method.
   View Comment on GitHub

3. Comment by 11Czh (March 5, 2025):

   Hello, regarding the ThinkRAG open-source framework, I want to fine-tune the model to test the results before and after fine-tuning. I want to use our framework to process input questions in batches and generate answers to compare the results before and after fine-tuning. I would like to ask if this framework can achieve this, not just for front-end Q&A. Looking forward to your reply.
   View Comment on GitHub

4. Comment by wzdavid (March 10, 2025):

   ThinkRAG is built on LlamaIndex. LlamaIndex has an evaluation module that can evaluate the effectiveness of RAG, which might suit your scenario, but ThinkRAG has not yet utilized it.
   Documentation Link
   View Comment on GitHub

---