I think it's my go-to 8b now.
I'd have to switch back to compare but I think I like it even more than my previous go-to of dark planet titan 12b. Strikes me as on-par with unslop nemo 12b but I only used it for a little bit before this came out. Hits far above its weight, gives me hope for us ramlets. I had some issues early on but that was surely user error.
Thank you Sicarius.
Indeed, one of my main motives is to make AI as accessible as possible :)
Very glad to hear you enjoy it!
@SicariusSicariiStuff Decent finetune, Sir. Highest score on my benchmark of all the Llama 3 8B variants I've tried.
This is the best model for me! I can only use models upto 8B due to my 6gb vram limitations. It stayed true to my system prompt, it understands the context very well. It can mess up once during a single response but the next line within the same response it somehow corrects itself.
Last thing, I am using your recommended settings for Roleplay text preset, default Llama 3 instruct: context and instruct templates.
The custom stopping strings "\n", "\nYou:"
Is there anything I am messing up? I am new to local models. So I have very less knowledge.
You could also use the ST config that can be downloaded for later models, if you're using ST.
Glad you enjoy it 🙂