Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Using your model in Home Assistant #64

Open
lordratner opened this issue Dec 26, 2024 · 4 comments
Open

Using your model in Home Assistant #64

lordratner opened this issue Dec 26, 2024 · 4 comments

Comments

@lordratner
Copy link

Hi,

I know this isn't the right repository exactly, but I couldn't find one for your assist-llm model on ollama.

I'm not able to use your Template in the "instructions" section of the Assist pipeline setup. It doesn't thrown an error, it just doesn't save. Or close. Nothing happens.

I'm attempting to use it through the ollama integration. Perhaps that's my problem? I was able to load the model fine, but the template is where I'm stuck. Do you have advice for running your model? I'm trying to find something that works locally, and coming up short.

Thanks!

@allenporter
Copy link
Owner

This is the right place, thanks -- (this repo does have the code for training assist-llm).

I'm not sure I understand what you're doing and running into a problem with. I would suggest breaking this down:

How about you try getting ollama integration working with llama 3.1? assist-llm does not yet provide meaningful improvements over it.

@lordratner
Copy link
Author

Oh, ok. I'm able to get llama3.1 and 3.2 working no issues.

When you se up a conversation agent in HA, it asks for "Instructions" which is the prompt that feeds the available entities, areas, and devices to the model. What I'm looking for is a set of instructions that works well with llama (or any other model I happen to be using).

I get a lot of problems where the agent is asking for the area or exact device I'm referencing. None of these issues exist when using one of the big three cloud AIs.

@allenporter
Copy link
Owner

allenporter commented Dec 26, 2024

I recommend using the default values and then also make sure to enable tools for Assist if you haven't already.

If you have a specific failure where it's not working we can work on improving the quality for it by default. Something simple like that should "just work" with no other work on your part.

@allenporter
Copy link
Owner

As an example here are some cases that use an area for targeting a cover with the default prompt:

-- should just work.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants