license: apache-2.0
language:
- en
model:
- MRockatansky/Cogidonia-v2-24B
pipeline_tag: text-generation
tags:
- text-generation-inference
Presenting Cogidonia v2!
Cogidonia v2 is a fine-tuned reasoning model. It is based on Cogidonia, a merge of two of TheDrummer's models, Precog-24B-v1, and MuXodious/Magidonia-24B-v4.3-absolute-heresy: https://huggingface.co/MRockatansky/Cogidonia-24B
I thought Cogidonia turned out really well for creative writing and so decided to also fine-tune it on a set of short stories and roleplays with various genres represented. The result is this model, Cogidonia v2, a model designed for creative writing which excels at historical and fantasy settings.
Chat template is mistral-v7-tekken and I use these settings:
Temp: 1
Top P: .95
Min P: 0.025
Top K: 40
Adaptive P: .5
Decay: .9
Uses <think> or <thinking> for reasoning. The reasoning on this model is short and doesn't consume a lot of tokens. The model will simply draft a brief synopsis of what it intends to write and then begins generating its response.
imatrix quants here: https://huggingface.co/MRockatansky/Cogidonia-v2-24B-imatrix-GGUF