|
|
--- |
|
|
base_model: |
|
|
- Gryphe/Pantheon-RP-1.8-24b-Small-3.1 |
|
|
- mrfakename/mistral-small-3.1-24b-instruct-2503-hf |
|
|
- OddTheGreat/Apparatus_24B |
|
|
library_name: transformers |
|
|
tags: |
|
|
- mergekit |
|
|
- merge |
|
|
- roleplay |
|
|
- creative |
|
|
language: |
|
|
- en |
|
|
- ru |
|
|
--- |
|
|
# Core |
|
|
|
|
|
This is a merge of pretrained language models |
|
|
|
|
|
With new Mistral recently released, and being slightly better than it predecessor, i wanted to update Apparatus. |
|
|
|
|
|
Also, i tested new pantheon, and i like how it mimic human expressions and mannerisms, so idea of new merge was born. |
|
|
|
|
|
Goal of this merge is transfer Apparatus to new Mistral 3.1, and enhance it dialogue capabilities while preserving it stability and ru performance. |
|
|
|
|
|
It seems that i succeed. Model is smart enough, instruction-following, decently creative and stable. |
|
|
|
|
|
tested on 250 answers, narration is really good, dialogues too, swipes make difference. |
|
|
|
|
|
Russian performance still here, no problems with full ru cards and partly translatd ones. |
|
|
|
|
|
Qvink Memory seems to break something. With it on, replies becomes much shorter. |
|
|
|
|
|
Better use prebuilt V7 format in ST, but chatML also work fine. T1.01 XTC 0.1 0.1 |
|
|
|
|
|
Is it better than Apparatus? i cannot tell, so your feedback is apperciated. |
|
|
|
|
|
P.S this model not include vision, but as soon i figure how to merge it in, i will update it. |
