LatitudeGames/Equinox-31B · Hugging Face

Posted by jacek2023@reddit | LocalLLaMA | View on Reddit | 28 comments

new model from LatitudeGames - Gemma 31B finetune

https://huggingface.co/LatitudeGames/Equinox-31B-GGUF

Equinox draws its name from the balance between extremes. Trained on a balanced blend of Wayfarer 2's unforgiving dark adventures and Hearthfire's quiet slice-of-life storytelling, Equinox is equally at home in perilous dungeons and candlelit conversations.

If you want to easily try this model, you can do so at https://aidungeon.com. Note that Equinox requires a subscription to use.

We plan to continue improving and open-sourcing similar models, so please share any and all feedback on how we can improve model behavior. Below we share more details on how Equinox was created.

[-]

Genebra_Checklist@reddit

Hey guys, kind off topic, but is there any Gemma 4 E4B or 26B A4B finnetune for creative writing and such?

[-]

GrizzyBurr_@reddit

ZeroFata has one on huggingface. zerofata/G4-MeroMero-26B-A4B

[-]

Borkato@reddit

MeroMero lowkey sucked for me 😭 im trying mero-artemis next, but so far i like Artemis and Skyfall better (tho Skyfall is mistral)

[-]

IrisColt@reddit

You need maybe the heretically ablated version. It took the crown in all my tests.

[-]

IrisColt@reddit

Thanks!!!

[-]

IrisColt@reddit

Gemma 31B finetune

I'm sold! Does it keep its vision capabilities?

[-]

VoiceApprehensive893@reddit

ai dungeon 💀

[-]

tableball35@reddit

Legit how I got started on the LLM road in like 2018-2019

[-]

10minOfNamingMyAcc@reddit

Same here. I was looking for alternatives to AI Dungeon and came across Tavern AI, which was later "replaced" by SillyTavern, and I’m still using that to this day. It feels kind of bad in a way to use their models locally now lol

[-]

Gryphe@reddit

It's still so surreal to be casually browsing Reddit and stumble on this, lol. Gemma was a pain to train - VRAM hungry, slow as heck, squiggly lines all over the place at the start, but as always I did the best I could. Considering this isn't my familiar Mistral architecture any and all feedback is very, very welcome!

[-]

Borkato@reddit

Holy shit it’s THE gryphe

[-]

seamonn@reddit

How useful is this finetune as an assistant?

[-]

jazir55@reddit

If you want to easily try this model, you can do so at https://aidungeon.com. Note that Equinox requires a subscription to use.

I'm not sure if we're operating on the same definition of "try".

[-]

seamonn@reddit

You can try with the help of your Credit Card.

[-]

Succubus-Empress@reddit

Can my tablet card run it?

[-]

kaisurniwurer@reddit

Try, as in the model the model. Just run it locally.

IQ4_XS available, awesome! Thanks!

Equinox on the aidungeon requires subscription.

[-]

a_beautiful_rhind@reddit

I liked wayfarer, hopefully they improved gemma.

[-]

jacek2023@reddit (OP)

What I think is interesting is that two years ago, the “most capable creative writing” arch was LLaMA 70B, and now it’s Gemma 31B. So something that was not really available to most home users (70B Q4 is still 35GB of VRAM) is now more accessible. And in those two years, the software has improved, so it’s actually faster on the same hardware.

[-]

kaisurniwurer@reddit

Pure writing capability, Llama 70B still holds. Though it has been a while since I used it because context is only so-so, so there might be some tinted glasses effect going on.

Maybe I'll try to check it out again.

[-]