SkunkApe-14b / README.md
MarsupialAI's picture
Create README.md
8a5d99f verified
|
raw
history blame
1.71 kB
metadata
license: cc-by-nc-4.0
language:
  - en
tags:
  - solar

Skunk Ape 14b

image/jpeg

This version performs substantially better than the 16b version.

This model is a merge of three self-merged Solar-based models in a 14b (64 layer) configuration. The result of this "frankenmerge" is a medium-sized model that contains what I consider to be the best of the solar finetunes.

Mergefuel:

  • Sao10K/Fimbulvetr-11B-v2
  • Sao10K/Solstice-11B-v1
  • TheDrummer/Moistral-11B-v1

This model is uncensored and capable of generating objectionable material. However, it is not an explicitely-NSFW model, and it has never "gone rogue" and tried to insert NSFW content into SFW prompts in my experience. As with any LLM, no factual claims made by the model should be taken at face value. You know that boilerplate safety disclaimer that most professional models have? Assume this has it too. This model is for entertainment purposes only.

iMatrix GGUFs: https://huggingface.co/MarsupialAI/SkunkApe-14b_iMatrix_GGUF

Sample output

{{[INPUT]}}
Write a detailed and humorous story about a cute and fluffy bunny that goes to a Gwar concert.
{{[OUTPUT]}}

<<<This goes on for a while.  See sample.txt for full output>>>

Prompt format

Prefers alpaca.

Weird merge fuckery

According to Toasty Pigeon, FM, Akai, and probably others on the KAI discord, this merge method works better than a normal stacked merge.
I don't pretend to understand why, but the PPL improvement (7.2 for this model vs 7.65 for the 16b @ Q4km) indicates that they're right.
See recipe.txt for all the alchemy.