Hi, I am just starting out with running a local language model on my ollama instance. I got everything set up, but when I have my conversation agent set to "control home assistant", I get an error when saying anything to it, "unexpected error during intent recognition". I am using gemma1.3b. am I required to set up custom sentences for it to work at all? Or is my model just failing to tell home assistant what to do?
#Ollama integration, unexpected error during intent recognition
abstract torrent
raw shard
It's very unlikely that you will have good results with 3B model for control. 7B also don't work with that. The bare minimum is something like qwen 14B, and still not that good.
abstract torrent
Oh wow okay
I found that my problem was that the model didn't support tools, whatever that is. I've switched to a llama 1b, and now it's just outputting json and not actually controlling any devices even though I have that enabled, and only a few exposed
I guess it's just not gonna happen with me running this on CPU lol