_np is not defined error after trying to index

[felixzsh]

❯ ccc init
Created project settings: /home/felix/Work/mcps/ate-framework-core/.cocoindex_code/settings.yml
You can edit the settings files to customize indexing behavior.
Run ccc index to build the index.

❯ ccc index
Project: /home/felix/Work/mcps/ate-framework-core
Indexing failed: name '_np' is not defined

---

i uninstalled my cocoindex because it was just too old and i didnt even hadd the ccc cli so followed the readme and now this doesnt work .... weird thing is that, i had a .cocoindex_code in that proyect (created with my previous cocoindex) at first i didnt delete that and ran ccc init and ccc index and everything seemed to work but then i just decided to better delete .cocoindex_code from the proyecto to start fresh, ran ccc init and ccc index and got that error

[georgeh0]

Hi @felixzsh , thanks for reporting the issue.

Can you try this to see if works:

ccc daemon restart

If not, can you check the file of ~/.cocoindex_code/daemon.log to see if there's any detailed stack trace for the Indexing failed: name '_np' is not defined error? Thanks!

[hxhieu]

I encountered the same issue and it looks like because of the new upstream CocoIndex SDK. It started happening on v0.2.22 when we bumped the SDK to 1.0.0a43.
For now I have to pin the upstream SDK to the earlier version for CCC to run.

[felixzsh]

in my case it was solved by changing the embedding model from nomic to snowflake artic, however i used to be able to use nomic

[hughbidone]

For me all built-in sentence_transformers plus embeddinggema are failing. They are fine if using CCC 0.2.11 right now.
I have also noticed first init/clean indexing is fine, only broken after that when calling ccc index again

@georgeh0 there is no logs, daemon triggering code not even run and it just crashes at "import" I guess - I am not Python expert though.

Could we please have a look team? Pretty much we cannot do incremental indexing now, have to start from scratch when adding new stuffs to the index. And when using 0.2.11 it is pulling the whole CUDA suite with 5GB+ on the container side while we are not using those built-in models.

Thanks.

[hughbidone]

It seems latest build yesterdat 0.2.28 fixed it for me with the stable SDK version of 1.0.0. Excellent!
Thanks!