---
library_name: transformers
tags:
- unsloth
- trl
- sft
- not-for-all-audiences
license: cc-by-nc-2.0
---

```
  e88 88e                               d8     
 d888 888b  8888 8888  ,"Y88b 888 8e   d88     
C8888 8888D 8888 8888 "8" 888 888 88b d88888   
 Y888 888P  Y888 888P ,ee 888 888 888  888     
  "88 88"    "88 88"  "88 888 888 888  888     
      b                                        
      8b,                                      
 
  e88'Y88                  d8           888    
 d888  'Y  ,"Y88b 888,8,  d88    ,e e,  888    
C8888     "8" 888 888 "  d88888 d88 88b 888    
 Y888  ,d ,ee 888 888     888   888   , 888    
  "88,d88 "88 888 888     888    "YeeP" 888    
                                               
PROUDLY PRESENTS         
```
# MN-12B-Tarsus-iMat-GGUF

Quantized with love from fp16.

Original model author: [envoid](https://huggingface.co/envoid/)

* Importance Matrix calculated using [groups_merged.txt](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384) in 92 chunks, n_ctx=512, and fp16 precision weights

Original model README [here](https://huggingface.co/Envoid/MN-12B-Tarsus/) and below:

## CAUTION: This model was finetuned on a corpus that includes adult content and may produce mature content without warning.
![](https://files.catbox.moe/1k5ama.jpg)

# MN-12B-Tarsus

Is a full-weight finetune of [mistralai/Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407)

Which underwent several intermediate steps.

This finetune was made with chatting/roleplaying via SillyTavern in mind and thus all of the testing was done there, with the goals being to:

-Reduce shiver-slop

-Make the model more conversationally proactive

-Give it a more human-like output (i.e. less gratuitous purple prose)

-Reducing overall positivity bias

It still responds well to Mistral-Instruct formatting.

The results are imperfect and its assistant capabilities suffered somewhat as a result but in quick testing it definitely seems to have achieved all of the goals to varying degrees. 

It sometimes fumbles with tokens in odd places so it's certainly not perfect. Possibly best used as merge-fodder.

Trained using [qlora-pipe](https://github.com/tdrussell/qlora-pipe)