Nimbus-Miqu-v0.1-70B Should Be Quantizable Now
I made a new Miqu merge, I saw it in your queue but due to terrible upload speeds I had to upload 1 .safetensors file at a time over days, so it would have failed 2 days ago.
Please let me know if you can still do GGUF quants for this model.
Ha! I saw and made a background job that would download new files every hour so I can immediately start when I see it's there, that's how stoked I am for a new miqu model :-) Will finally something be able to beat quartetanemoi? I need to find out :)
Anyway, it converted without issues, so unless something is drastically wrong with the model weights, quants will be there within a day. Cheers!
Ha! I saw and made a background job that would download new files every hour so I can immediately start when I see it's there, that's how stoked I am for a new miqu model :-) Will finally something be able to beat quartetanemoi? I need to find out :)
Anyway, it converted without issues, so unless something is drastically wrong with the model weights, quants will be there within a day. Cheers!
Let's hope so, my main aim was to merge some of the positivity bias out of Midnight Miqu without losing the creativity.
No pressure :)
Was talking to it a few hours today, and I am impressed. Follows commands pretty well (compared to midnight miqu), is not too moist, and does not seem to have suffered elsewise. And this was with an IQ3_XXS (because it was the one available at the time). Well done!
Was talking to it a few hours today, and I am impressed. Follows commands pretty well (compared to midnight miqu), is not too moist, and does not seem to have suffered elsewise. And this was with an IQ3_XXS (because it was the one available at the time). Well done!
Thanks, yeah it's definitely showing promise. I have seen some issues especially with Claude-formatted character cards (which Midnight handles pretty well).
ChatML has given me better results than Alpaca which I definitely didn't expect.
Let me know what you run into!
So, I was playing around with it for the last few days, and I find it way better than midnight miqu or moist miqu. Didn't even bother with upgrading the IQ3_XXS quant (right, thanks for reminding me :).
Congrats, this is the first quartetanemoi alternative I have seen yet (but then, I have rather specific requirements, so I have no clue how well it performs chatting or rp'ing - but those things tend to be easier, so maybe it's even better doing that :).