43 lines
1.3 KiB
Markdown
43 lines
1.3 KiB
Markdown
|
---
|
||
|
|
license: apache-2.0
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
base_model:
|
||
|
|
- MRockatansky/Cogidonia-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
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|