Chunk/context limit broken or too limited?

[mhbates]

Which version of Nextcloud are you using?

31

Which version of PHP context_chat are you using?

4.5.0

Which version of backend context_chat are you using?

4.5.0

Nextcloud deployment method?

docker-compose

Describe the Bug

I noticed with Context Chat that I have to give it very selective context for it to work as expected - as in pointing it at the exact subfolder (or one folder up from that) for it to find the file I'm asking about. If I query from higher than that, it seems to only include a subset of my data in the context, and thus doesn't find the file. I also see the following errors in the log and I'm not sure if these are related, but I'm guessing not.

decode: cannot decode batches with this context (calling encode() instead) init: embeddings required but some input tokens were not marked as outputs -> overriding

To Reproduce

Ask Context Chat a question about a file's contents from a higher folder level, aka, make it have to look through more data to find what you're asking about.

PHP logs (Warning these might contain sensitive information)

No response

Ex-App logs (Warning these might contain sensitive information)

No response

Server logs (if applicable)

No response

[kyteinsky]

hello,

Ask Context Chat a question about a file's contents from a higher folder level, aka, make it have to look through more data to find what you're asking about.

there's a feature request to make it configurable: nextcloud/context_chat#134
for now it is 20 chunks

I also see the following errors in the log and I'm not sure if these are related, but I'm guessing not.

they might be, it is tracked here: #218

it would be nice if you could provide a sample set of docs and search queries to compare our improvements since our current CI tests have a lot of docs and our multilingual queries easily find the relevant docs.

[mhbates]

I'm not able to provide sample docs, but I can give a sense of the structure and size of data, specifically number of files:

Query: "Give me the license plate number of the Toyota Prius"

• Documents (root)
  -- Folder A (2220 files)
  --- Car (141 files)
  ---- Toyota (11 files)
  ----- ToyotaPriusLicensePlate.txt

I need to point Context Chat at the "Toyota" folder for it to succeed. Won't work at the "Car" level. Hope this helps.