#ooooolllaaamaa
loud vector
add service - enter host IP - then my screenshotted message shows up
shell kayak
oh right
that is now the big benefit of sub entries
why would you need to set up the whole service over again?
The other button is to add a new conversation agent for the existing entry
loud vector
Yeah I get that - the conversation agent button doesnt have a place to specify which model to use within my ollama host
unless its the name input that determines the model- if so it should probably have a section below stating that
shell kayak
@snow hearth heh?
oh i see in the config flow that we indeed ask the model in the config flow part
Can you open an issue so we can track this?
We either need to move the model part to the conversation agent setup, or change the things we validate on
loud vector
Happy to help but I've never done this before - is there a doc/steps I can follow?
shell kayak
It's a form and I think you can just explain that the config entry validates on host, and not host + model
loud vector
will do, thanks!
shell kayak
Thanks ❤️
shell kayak
Thanks! I think I understand the issue, I am now just checking with Paulus what we should do with this one
snow hearth
Yeah interesting!
I guess for Google we allow model per sub entry
So we should do here too
shell kayak
That probably makes the most sense
snow hearth
Although Ollama works slightly different in that it holds one model in memory
We set the config entry name ti model. Should turn that generic too then
shell kayak
I believe it might be merged for the migration
But what about that one model in memory, what is the best way to proceed here
does that affect anything?
ok I should really start packing my bag lol
loud vector
for what its worth - prior to this beta you could specify multiple models by going through the config flow again
I think you can have multiple models loaded into memory with the "keepalive -1" parameter memory space permitting
snow hearth
Yeah you can still make multiple config entries
If someone can run two models in memory they usually run just 1 bigger model 🙂
shell kayak
Okay so then we would move the model selection to the sub entries?
I'm not sure if I have time before Wednesday as I'm leaving for Luxembourg tomorrow
@snow hearth Should users be able to change the model for a sub entry?
snow hearth
So with Ollama we don't recommend it, but I guess we do allow it also for Google etc
the thing with Ollama is that we don't have a preferred model, we do show downloaded model
let me loop in @orchid moss
shell kayak
Uhh, why does ollama use the user step for multiple things
orchid moss
Is the spirit of sub-entries more for the service or the service and the model? I thought just the service, since the original idea was to avoid duplicate API keys. But I can see an argument for both.