Find answers from the community

Updated 3 months ago

llama pack customization.

Hi, I'm trying to use https://llamahub.ai/l/llama_packs-dense_x_retrieval?from=llama_packs
I wonder the llama_packs is a ready to use package, or is it more like a template that we can build on? For example I like to add storage context for vector store index, but not sure if I should contribute it to the llama_pack or just add the code locally.

7 comments

rravitheja

https://github.com/run-llama/llama_index/blob/main/docs/examples/llama_hub/llama_packs_example.ipynb - You can customize the pack locally. You can check here about customization

SSan Nguyen

hi @Logan M , sorry for bothering. I just try the dense_x_retrieval llama pack. I got an error where parsing the text to list[str] is not correct. I'm not sure the purpose of parsing the response to a list[str] instead of a str. Could you elaborate a bit? The parsing code seems prone to errors.

Attachment

LLogan M

The parsing code is so complicated haha

I was copying the authors prompt though

LLogan M

If you have a better way to parse it, I welcome a PR. But it seemed to work well in my testing with openai

SSan Nguyen

what will happen if I just use the output for the proposition instead of parsing it to list[str]?

LLogan M

Then there would be a single node representing all propsitions instead of a node for each proposition

LLogan M

I imagine it would have some impact on the quality of retrieval, but tbh probably not too noticable

Add a reply