Thank you

[akavel]

Shit; this thing actually works!...

I just started trying it - and it's a completely different league than the minuet-ai.nvim one I tried before with ollama. That one was trying to do some completion, sure, but it was mostly useless, only rarely anything was actually useful, so that when it was, it usually surprised me.

Here, wow - I just accepted quite a few completions in Rust, and they were what I wanted, and compiled, and worked. Sure, nothing super ambitious for the time being - but for a glorified autocomplete, this one already shows up to be quite a nice and usefully glorified one!

And I'm using just the 7B model! (The suggested llama-server --fim-qwen-7b-default.) I'll have to try also codellama 7b at some point, it felt a bit more convincing (though slower) in the old minuet+ollama config; though admittedly the qwen2.5-coder I tried there was only a q4_K_M, not a full q8_0.

Anyway, I'm heavily impressed, and thank you! I'm a newbie to all this LLM stuff, and this is starting to get interesting...

[natedemoss]

glad you enjoy it!