Coding model demo

[srid]

We already have a chatbot example: https://community.flake.parts/services-flake/llm

A coding agent example would be interesting as well:

https://ollama.com/blog/coding-models

https://news.ycombinator.com/item?id=45601834

[shivaraj-bh]

https://ollama.com/blog/coding-models

They have local and cloud variant. The latter requires authentication.

For the cloud variant, we will have to create an additional process that will open the browser for the user to authenticate and keep polling until the response changes from

{"error":"unauthorized","signin_url":"https://ollama.com/connect?name=shivaraj-mac\u0026key=c3NoLWVkMjU1MTkgQ[...]WM"}%  

to success.

[shivaraj-bh]

The smallest qwen3-coder from the blog that can run locally is 19GB and when I try to download that model, the speed drops from 50MB/s to 1-2MB/s for the last ~5GB.

I tried the previous gen qwen2.5-coder:latest (4.7GB) and it is bad for the example prompt from the blog:

(generated html is just a score counter in the top-left corner)