---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3.5-9B
base_model_relation: finetune
pipeline_tag: text-generation
tags:
- RP
- ERP
- Experimental
- creative
- roleplay
- writer
- creative writing
- high rank
- potentially unstable
library_name: transformers
---

One of the many testing version for my upcoming Nyx-RP-9B-Instruct-2608-v2, this one had its stability improved after I carefully analyzed the first versions.

Training details:

## Method: LoRA

* Rank: 226
* Epochs: 2(decreased from previous 3, which did seem to be the breaking point)

## Hardware used for training(I have patience):

* RTX 5070 Ti
* R7800X3D + 32gb DDR5

## Dataset detail:

Its a fairly new dataset which I gathered from the best RP logs I had on my hands, it included a lot of explicit long turn conversations so keep in mind the reduced refusals.