PydanticAI using Ollama (llama3.2) running locally (opens in new tab)

(github.com)

2 pointsscolvin1y ago3 comments

3 comments

So cool! I wonder what the weakest model that can still call functions and such is?

I don't have anything more powerful than an i5 other than my phone, and a lot of interesting applications like home automation really need to be local-first for reliability.

0.5b to 1b models seem to have issues with even pretty basic reasoning and question answering, but maybe I'm just Doing It Wrong.

scolvinOP1y ago

See https://github.com/pydantic/pydantic-ai/issues/112 people have tried quite a few models.

Llama3.2 worked well and used <2gb ram

eternityforest1y ago

Edit: Gemma2 2B is very slow but it is able to do some basic tasks

j / k navigate · click thread line to collapse

3 comments

eternityforest1y ago

So cool! I wonder what the weakest model that can still call functions and such is?

I don't have anything more powerful than an i5 other than my phone, and a lot of interesting applications like home automation really need to be local-first for reliability.

0.5b to 1b models seem to have issues with even pretty basic reasoning and question answering, but maybe I'm just Doing It Wrong.

scolvinOP1y ago

See https://github.com/pydantic/pydantic-ai/issues/112 people have tried quite a few models.

Llama3.2 worked well and used <2gb ram

eternityforest1y ago

Edit: Gemma2 2B is very slow but it is able to do some basic tasks

j / k navigate · click thread line to collapse