"Help modifying prompt.py without breaking the model"

[Franco-vrtx]

Hello LightRAG community!
I’d like to ask if you have any guide or documentation on how to modify prompt.py without breaking everything. I tried changing some things related to the entities to extract, and when making queries, the model responded with something like '''csv.... ''' and nonsensical data.

I think the most important thing is knowing how to properly structure the extraction prompts. Any tips are welcome! Thanks!

[BireleyX]

I customized prompt.py for the kind of documents I use and I saw a marked improvement with the generated entities and relationships.
What I did was to add my own examples based on actual sample data from my documents. I also customized the response prompts to generate references on modes other than "mix" (on an earlier LightRag version that didn't support it yet).

it could be some typo, if you modified the example prompts.

[ntsarb]

I customized prompt.py for the kind of documents I use and I saw a marked improvement with the generated entities and relationships. What I did was to add my own examples based on actual sample data from my documents. I also customized the response prompts to generate references on modes other than "mix" (on an earlier LightRag version that didn't support it yet).

it could be some typo, if you modified the example prompts.

When I modify PROMPTS["DEFAULT_ENTITY_TYPES"] in the prompt.py file, and run lightrag-server, the changes are ignored. I still get the default entity types. I deleted the corresponding file in pychace just in case, still nothing, which I find weird. I must be missing something.

Can you please provide more info?