I am using the multi-doc

SSimon 📐🛠

I am using the multi-doc_agents-v1 google colab notebook and tweaking it with my own data. I ran it fine with the original data set (llama index html files) but with my own data I rely of the tool names to have numbers and periods which seem to not gel with openai chat. Is there a way I can use numbers as they are essential to the hierarchical nature of these documents?

Plain Text

/usr/local/lib/python3.10/dist-packages/openai/_base_client.py in _request(self, cast_to, options, remaining_retries, stream, stream_cls)
    991 
    992             log.debug("Re-raising status error")
--> 993             raise self._make_status_error_from_response(err.response) from None
    994 
    995         return self._process_response(

BadRequestError: Error code: 400 - {'error': {'message': "'tool_35.24.020 - Purposes of Commercial Zones_C-2 (Retail Commercial) zone.' does not match '^[a-zA-Z0-9_-]{1,64}$' - 'tools.0.function.name'", 'type': 'invalid_request_error', 'param': None, 'code': None}}

6 comments

LLogan M

Yea, tool names can only contain certain characters, need to clean that up

LLogan M

the regex there is telling you the allowed chars

LLogan M

ah I see you got it

SSimon 📐🛠

Just curious does it matter how unique the name of the tool is? More specifically, at query time is the tool name checked for semantic similarities with query BEFORE the top agent dives into the underlying SummaryIndex or / and VectorIndex?
For instance I have a data set with titles starting with

Plain Text

35.24.020

which correlate to

Plain Text

(Article.Chapter.Section)

which is useful as a reference number but not so useful semantically as it wont correlate with the query. I now have the number removed completely like

Plain Text

Purposes_of_Commercial_Zones_C-2_Retail_Commercial_zone

but could I leverage this number as well to better inform my top agent in picking the correct Index's to answer the query?

LLogan M

The tool should hopefully have a name that makes it indentifiable and also easy for the LLM to write

There isn't semantic similarity checks or anything unless you explicitly used a tool retriever

SSimon 📐🛠

okay great thanks!

Add a reply

Find answers from the community

I am using the multi-doc_agents-v1