library_name: transformers
tags:
- merged-model
- merge
- mergekit
- Uncensored
- vivid writing
- novel
- roleplay
- nsfw
- not for all audiences
base_model: - "petruhonk/Qwen3.8-9B-Distill-uncensored-heretic"
- "DavidAU/Qwen3.5-9B-The-Bradbury-F451-Pro-Writer-Uncensored-Heretic"
language: - zh
两个精心挑选的模型，合并出一个奇迹
Two handpicked models, merged into a miracle.
虽然这是我第一次合并模型，但我认为结果非常好。特别在中文nsfw、角色扮演、小说方面
Although this is my first time merging models, I think the results are excellent, especially for Chinese NSFW, role-playing, and novels.
源模型来自：
The source model is from:
petruhonk/Qwen3.8-9B-Distill-uncensored-heretic
DavidAU/Qwen3.5-9B-The-Bradbury-F451-Pro-Writer-Uncensored-Heretic
使用 kaggle 免费会话和 mergekit 开源工具完成
Completed using Kaggle's free sessions and the open-source tool Mergekit.
非常感谢他们！
Thank you so much!
merged
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the SLERP merge method.
Models Merged
The following models were included in the merge:
- /kaggle/temp/model_a
- /kaggle/temp/model_b
Configuration
The following YAML configuration was used to produce this model:
merge_method: slerp
models:
- model: /kaggle/temp/model_a
- model: /kaggle/temp/model_b
base_model: /kaggle/temp/model_a
parameters:
t:
- filter: self_attn
value: [0, 0.5, 0.3, 0.7, 1]
- filter: mlp
value: [1, 0.5, 0.7, 0.3, 0]
- value: 0.5
dtype: bfloat16
Merge recipe
SLERP merge (mergekit) combining two Qwen3.5-9B finetunes:
- petruhonk/Qwen3.8-9B-Distill-uncensored-heretic - Qwen3.8 full-parameter distillation into Qwen3.5-9B, focused on general intelligence (math / code / reasoning / tool use), heretic-uncensored. Used as the merge base.
- DavidAU/Qwen3.5-9B-The-Bradbury-F451-Pro-Writer-Uncensored-Heretic - creative-writing specialist (prose style), heretic-uncensored.
Layer-wise t schedule (t=0 -> base/general, t=1 -> writer):
- self_attn: [0, 0.5, 0.3, 0.7, 1]
- mlp: [1, 0.5, 0.7, 0.3, 0]
- all other tensors: 0.5
dtype: bfloat16. Early layers lean toward the distilled reasoning model while late attention
layers lean toward the writer, aiming to keep general intelligence and add creative-writing
strength. Full mergekit config: see merge_config.yaml.
Warning: parents are uncensored/ablated models; outputs are not safety-filtered. Not for all audiences.