Documentation for LLM training

Getting Started with n8n Docs & Tutorials
1

I am puzzled by the lack of comprehensive documentation for training LLM models. Why isn’t there a repository on GitHub where developers can access the latest versions in text format?

While there is a website (https://docs.n8n.io/), it does not provide the tools needed to train models effectively. Today, even advanced models like Claude Opus struggle to work well with N8N code. They lack the nuanced understanding required to handle complex tasks. Even when using the MCP, creating connections is fraught with issues.

I have spent considerable time setting up the MCP, yet the results have been unsatisfactory. Claude Opus is not trained to grasp the subtleties of the N8N code. Although it can generate JSON code for a bet, the overall performance is poor.

I apologize for expressing my frustration, but it is difficult to remain silent when obvious problems are not being addressed. A comprehensive documentation for RAG would greatly simplify development.

2

Hey @m_sav47 hope all is good. Welcome to the community.

When you say “for training LLM models”, what do you mean exactly? The term “model training” is heavily overloaded these days. Do you mean a structured description of every node, which could be fed to the external system to better understand the building blocks of N8N?

3

Greetings! Thanks for your reply. You’re absolutely right.

We need a structured description of each node. It can be transferred to an external system. This will help to better understand the building blocks of N8N. It is important that the description can be conveyed as text.

4

Yeah, I don’t believe anything like that exist, unless you want to feed the codebase of the nodes themself:

1 Like
5

Thank you! There are many files. I received my golden antelope, but the description is missing. I can’t merge them into one text file easily. :melting_face:

6

You could parse the codebase, build your own node description database, by feeding each node code to your coding LLM, providing context and asking to write a description for what this node is, what it does, how and when to use it based on the code.

You could then share your built database with the community :wink:…

7

Hello, Jabbson! I am writing to express my gratitude for your prompt response. I have taken it upon myself to teach others how to do this, and I have done so. I am aware that the documentation can quickly become outdated, and therefore it must be regularly updated. However, it is still better than nothing. :wink: