Spaces:
Configuration error
New base model
Well, my friends... Forgotten Abomination has become my go-to writing model. It's an amazing model.
Its only real weakness is the 12B version: it doesn't work particularly well with agent frameworks like Hermes or OpenClaw because it lacks proper tool calling and other agent-oriented capabilities.
The larger models, like Forgotten 31B, are phenomenal, but they're simply too large as dense models for most consumer hardware. I've also tried several of the new 26B MoE models based on Gemma like Dark-scarlet , and while they're good, none of them reach the writing quality of Forgotten Abomination or Forgotten safeword.
Then, this week, a miracle happened.
Bonsai 27B and Bonsai Ternary 27B were released.
And let me be very clear: Bonsai Ternary 27B has roughly the same footprint as Forgotten Abomination 12B, while Bonsai 27B can run in about 4 GB. That's absolutely insane.
So here's my proposal:
Would it be too much to ask for the Forgotten treatment to be applied to the Bonsai 27B family?
What can I do to help make Forgotten Bonsai Abomination a reality?
How?
We don't have the datasets for forgotten abomination and sleep deprived passed away.
So... Rock and hard place
WHAAAT?! Don't play with my heart like that!
Well, I'm definitely not an experienced model trainer, but I'll see what I can do. I'll try to find another model with great storytelling abilities and look into open-source datasets or interesting fine-tuning heretic/obliterated projects that could be used as a starting point.
Maybe I'll stumble upon something useful. I'd love to help somehow!